value per subscription dollar · curated plan terms← intelligence board

5dive Pulse · Token Maxxing

Which AI subscription gives you the most tokens per dollar?

We take what each plan lets you use in a month, price it at what those tokens would cost you through the API, and divide by what the plan costs. Higher means more for the same money.

Most providers do not publish a usage limit, so most of these numbers are estimates. Every row shows how it was worked out, what it rests on, and when we last checked it.

Value per subscription dollar

Individual plans, from the model providers themselves. This ranks value, not model quality — that is a separate question, on the intelligence board.

Each row is what a plan gives an agent that runs until the plan stops it. Open one for its range, sources and check date; rows we metered ourselves say so.

  1. 1

    Claude Pro

    $20/mo · Claude Opus 5

    ~20×

    Claude Pro

    rank #1 · Claude Opus 5

    ~20×value per $ (headline)
    Maxxed range
    20×
    Evidence
    Modelled · n=0
    Time-sensitive
    TEMPORARY, THROUGH 2026-09-13: Anthropic users report a +50% weekly-limit boost running to that date, so any Claude measurement taken inside that window overstates what the plan sustains afterwards. This row is MODELLED and does not incorporate the boost, and the community figures it cites were logged before it — but treat any Claude number you read anywhere this week, including your own, as inflated until the cut lands.
    Binding limit
    weekly + 5h windowA rolling 5-hour window plus a weekly ceiling on the newest model. Which one binds depends on whether the work is bursty or continuous, and neither is published as a number.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    High
    Throughput
    Low · high volatility
    List price
    $20/mo
    Token value
    $410–$410
    Sources
    [1][2]
    Last checked
    checked Sep 5

    Rolling 5h window + weekly cap, ~5× the free tier; absolute messages/window no longer published. Entry tier — the 1× reference for Anthropic. Modelled: Anthropic publishes no ceiling we can apply and we do not meter this plan, so the throughput is the generic workload band.

  2. 2

    ChatGPT Plus

    $20/mo · GPT-5.6 Sol

    ~17×

    ChatGPT Plus

    rank #2 · GPT-5.6 Sol

    ~17×value per $ (headline)
    Maxxed range
    17×
    Observed
    ~17× · $348/mo API-equivalentbeside the modelled headline, not in place of it
    Evidence
    Observed · n=1 · seen Sep 7
    Surface ranked
    Codex (coding agent)This row ranks OpenAI's CODEX surface only, and the five-hour message ranges it publishes there. OpenAI meters Chat, Codex and Work on SEPARATE allowances, which cannot be added, averaged or blended into one number — doing that would invent a quota OpenAI does not publish. Codex is the surface this board is about (the same scope test that removed SuperGrok Lite and Google AI Plus), so it is the one ranked, and the chat allowances below are recorded as OUT OF SCOPE rather than folded in. ATTESTED, NOT READ: the chat figures come from the external audit's reading of help.openai.com articles that return HTTP 403 to every automated read from this host, so we publish them as somebody else's reading, not ours. On chat, Plus is the entry tier; the tie the audit breaks is between the two Pro tiers.
    Binding limit
    weekly + 5h windowOpenAI publishes local-message estimates per FIVE-HOUR period, per model and per tier (Plus / Pro 5x / Pro 20x), and states that weekly limits may also apply — so the window structure IS published, contrary to what this row said before. They are RANGES, explicitly disclaimed as “not fixed message limits” with the live numbers held in the usage dashboard, so the shape is published even though the quota is not.
    Maxxed ceiling
    metered cycleDerivation 1, and the only LIVE one on this board: a cycle we metered ourselves against OpenAI’s own quota gauge. An 18-minute agent task on the Codex CLI drew 14,016,606 tokens (98.2% cache reads) and the ChatGPT analytics page moved from 100% to 91% of the weekly limit remaining, so that one task was 9% of a Plus week. Published in DOLLARS at OpenAI’s own published rates — $80/week, ~$348/mo — because the measured cycle carries its own token mix and must never be re-valued through the board’s modelled one. NO TOKEN CEILING IS PUBLISHED FOR THIS ROW ON PURPOSE: extrapolating the tokens gives ~155.7M/week on the raw count and ~3.17M/week on fresh tokens only, 49x apart, because OpenAI meters this allowance in credits and requests rather than tokens. The dollar value of the cycle does not depend on that choice, which is why it is the number we sign.
    Plan terms
    Med
    Throughput
    Med · high volatility
    List price
    $20/mo
    Token value
    $348–$348
    Last checked
    checked Sep 6

    Standard text unlimited; only reasoning/tools metered, caps published only as per-five-hour message ranges on developers.openai.com/codex/pricing, disclaimed as not fixed. Entry paid tier — the 1× reference for OpenAI. Terms graded medium: openai.com and help.openai.com both refuse an automated read (HTTP 403), so the $20 price is cross-checked against secondary trackers, not read off the vendor page. THE RANK NO LONGER COMES FROM THAT MODEL: since 2026-09-07 this row is ranked on a cycle we metered on our own Plus account against OpenAI’s own weekly gauge — see Observed above. The published ranges stay on the row as what the vendor says, not as what the row is scored on.

  3. 3

    Claude Max (5×)

    $100/mo · Claude Opus 5

    ~17×

    Claude Max (5×)

    rank #3 · Claude Opus 5

    ~17×value per $ (headline)
    Maxxed range
    14–20×Headline is the centre: both ends are quota-saturated and equally likely.
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Time-sensitive
    TEMPORARY, THROUGH 2026-09-13: Anthropic users report a +50% weekly-limit boost running to that date, so any Claude measurement taken inside that window overstates what the plan sustains afterwards. This row is MODELLED and does not incorporate the boost, and the community figures it cites were logged before it — but treat any Claude number you read anywhere this week, including your own, as inflated until the cut lands.
    Binding limit
    weekly + 5h windowA rolling 5-hour window plus a weekly ceiling on the newest model. Which one binds depends on whether the work is bursty or continuous, and neither is published as a number.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). This row tops out at the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”. This row keeps its BAND rather than collapsing to that ceiling, because the band is not a workload guess — it is how much this tier delivers relative to Claude Pro, which is observed with real spread and which saturating the plan does not resolve.
    Plan terms
    High
    Throughput
    Low · high volatility
    List price
    $100/mo
    Token value
    $1,435–$2,049
    Sources
    [1][2]
    Last checked
    checked Sep 5

    “5×” is the 5-hour-session figure, not monthly; a June 2026 class action alleges ~3.5× Pro in practice. Modelled at 3.5× Pro at the floor and at the advertised 5× at the ceiling. Absolute caps unpublished. Community-observed rather than vendor-derived: the 3.5× floor comes from third-party allegation and user logs, not from anything Anthropic publishes.

  4. 4

    ChatGPT Pro (5×)

    $100/mo · GPT-5.6 Sol

    ~16×

    ChatGPT Pro (5×)

    rank #4 · GPT-5.6 Sol

    ~16×value per $ (headline)
    Maxxed range
    16×
    Evidence
    Modelled · n=0
    Surface ranked
    Codex (coding agent)This row ranks OpenAI's CODEX surface only, and the five-hour message ranges it publishes there. OpenAI meters Chat, Codex and Work on SEPARATE allowances, which cannot be added, averaged or blended into one number — doing that would invent a quota OpenAI does not publish. Codex is the surface this board is about (the same scope test that removed SuperGrok Lite and Google AI Plus), so it is the one ranked, and the chat allowances below are recorded as OUT OF SCOPE rather than folded in. ATTESTED, NOT READ: the chat figures come from the external audit's reading of help.openai.com articles that return HTTP 403 to every automated read from this host, so we publish them as somebody else's reading, not ours. ON CHAT, THE TIERS ARE NOT TIED: the audit reports Pro $100 at 50 GPT-6 Pro messages per week against Pro $200's 200 per week — 4x the weekly allowance for 2x the price, so on THAT surface $200 is the better value per dollar, the reverse of a tie. It does not move this row, because this row is not ranking that surface.
    Binding limit
    weekly + 5h windowOpenAI publishes local-message estimates per FIVE-HOUR period, per model and per tier (Plus / Pro 5x / Pro 20x), and states that weekly limits may also apply — so the window structure IS published, contrary to what this row said before. They are RANGES, explicitly disclaimed as “not fixed message limits” with the live numbers held in the usage dashboard, so the shape is published even though the quota is not.
    Maxxed ceiling
    modelled (no quota published)OpenAI publishes per-five-hour message RANGES rather than a token quota, and disclaims them as not fixed, so there is still no cycle this board can saturate arithmetically (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $100/mo
    Token value
    $1,640–$1,640
    Tied tier
    unmeasured; tied by assumptionit prints the same rate as its other OpenAI tiers because neither is measured and the model scales the dearer one proportional to price
    Last checked
    checked Sep 6

    OpenAI sells Pro at two usage tiers, $100 (5× Plus) and $200 (20× Plus), same models and features — the price buys headroom only. Modelled proportional to price (5× Plus), not at the advertised 5× session label; they happen to coincide here. Terms graded medium: vendor pages return 403, cross-checked against secondary trackers.

  5. 5

    ChatGPT Pro (20×)

    $200/mo · GPT-5.6 Sol

    ~16×

    ChatGPT Pro (20×)

    rank #5 · GPT-5.6 Sol

    ~16×value per $ (headline)
    Maxxed range
    16×
    Evidence
    Modelled · n=0
    Surface ranked
    Codex (coding agent)This row ranks OpenAI's CODEX surface only, and the five-hour message ranges it publishes there. OpenAI meters Chat, Codex and Work on SEPARATE allowances, which cannot be added, averaged or blended into one number — doing that would invent a quota OpenAI does not publish. Codex is the surface this board is about (the same scope test that removed SuperGrok Lite and Google AI Plus), so it is the one ranked, and the chat allowances below are recorded as OUT OF SCOPE rather than folded in. ATTESTED, NOT READ: the chat figures come from the external audit's reading of help.openai.com articles that return HTTP 403 to every automated read from this host, so we publish them as somebody else's reading, not ours. ON CHAT, THE TIERS ARE NOT TIED: the audit reports this tier at 200 GPT-6 Pro messages per week (with GPT-5.6 Sol Pro at 170/day and both Pro models capped together at 200/day) against Pro $100's 50 per week — 4x the weekly allowance for 2x the price. On the chat surface this tier is therefore BETTER value per dollar than Pro $100, not tied with it; on the Codex surface ranked here, neither of them is measured at all.
    Binding limit
    weekly + 5h windowOpenAI publishes local-message estimates per FIVE-HOUR period, per model and per tier (Plus / Pro 5x / Pro 20x), and states that weekly limits may also apply — so the window structure IS published, contrary to what this row said before. They are RANGES, explicitly disclaimed as “not fixed message limits” with the live numbers held in the usage dashboard, so the shape is published even though the quota is not.
    Maxxed ceiling
    modelled (no quota published)OpenAI publishes per-five-hour message RANGES rather than a token quota, and disclaims them as not fixed, so there is still no cycle this board can saturate arithmetically (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $200/mo
    Token value
    $3,279–$3,279
    Tied tier
    unmeasured; tied by assumptionit prints the same rate as its other OpenAI tiers because neither is measured and the model scales the dearer one proportional to price
    Last checked
    checked Sep 6

    The advertised “20× Plus” is a per-FIVE-HOUR rate-limit figure, not a sustained monthly one: OpenAI’s Codex pricing page scales its five-hour message ranges 5× and 20× off Plus, then disclaims them as not fixed limits, and publishes no monthly quota at all. A disclaimed burst range is not a ceiling this board can rank on — the same test applied to Google’s unperiodised 20× — so this stays modelled proportional to price (10× Plus). Terms graded medium: vendor pages return 403, cross-checked against secondary trackers.

  6. 6

    Claude Max (20×)

    $200/mo · Claude Opus 5

    ~15×

    Claude Max (20×)

    rank #6 · Claude Opus 5

    ~15×value per $ (headline)
    Maxxed range
    11–22×Headline is the centre: both ends are quota-saturated and equally likely.
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Time-sensitive
    TEMPORARY, THROUGH 2026-09-13: Anthropic users report a +50% weekly-limit boost running to that date, so any Claude measurement taken inside that window overstates what the plan sustains afterwards. This row is MODELLED and does not incorporate the boost, and the community figures it cites were logged before it — but treat any Claude number you read anywhere this week, including your own, as inflated until the cut lands.
    Binding limit
    weekly + 5h windowA rolling 5-hour window plus a weekly ceiling on the newest model. Which one binds depends on whether the work is bursty or continuous, and neither is published as a number.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). This row tops out at the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”. This row keeps its BAND rather than collapsing to that ceiling, because the band is not a workload guess — it is how much this tier delivers relative to Claude Pro, which is observed with real spread and which saturating the plan does not resolve.
    Plan terms
    High
    Throughput
    Low · high volatility
    List price
    $200/mo
    Token value
    $2,152–$4,304
    Last checked
    checked Sep 6

    “20×” is Anthropic's 5-hour-SESSION figure, not a monthly one — the weekly cap is what binds a month, and 20× the burst does not buy 4× the week. This row used to model that gap as a fixed ~1.7× the 5× tier, which is one decimal of precision on a class-action allegation. It is now a RANGE, ~1.5–2.1× the 5× tier's sustained throughput, taken from the August 2026 community measurement of both tiers' weekly consumption — whose author later revised it upward, to “at best a little over 2×” — rather than from the lawsuit. Against the 5× tier's own 3.5–5× band that is ~5.25× Pro at the floor and ~10.5× at the ceiling. It still lands BELOW the 5× tier on value per dollar: 2× the price for well under 2× the sustained week. Absolute caps unpublished, and the community sources are read through the audit — reddit.com returns HTTP 403 to every automated read from this host.

  7. 7

    Google AI Pro

    $19.99/mo · Gemini 3 Pro

    ~9.5×

    Google AI Pro

    rank #7 · Gemini 3 Pro

    ~9.5×value per $ (headline)
    Maxxed range
    9.5×
    Evidence
    Modelled · n=0
    Binding limit
    weekly + 5h windowRolling 5-hour compute windows plus a longer ceiling, with each tier published as a usage RATIO of AI Pro rather than as tokens.
    Maxxed ceiling
    modelled (no quota published)The vendor publishes a usage RATIO between tiers but no absolute quota, so there is still no cycle to saturate (derivation 4). The ratio sets where this tier sits relative to its siblings; it does not say how much either of them is. Same upper-bound caveat as every modelled row.
    Plan terms
    High
    Throughput
    Low · medium volatility
    List price
    $19.99/mo
    Token value
    $190–$190
    Tied tier
    unmeasured; tied by assumptionit prints the same rate as its other Google tiers because neither is measured and the model scales the dearer one proportional to price
    Sources
    [1][2]
    Last checked
    checked Sep 6

    $19.99/mo (formerly Gemini Advanced). Pro anchors the 1× for Google. CORRECTION 2026-09-06 (second): this row briefly read the post-I/O multipliers (Pro 4× non-plan Gemini app usage, Ultra 5× and 20× Pro) as published ceilings and graded the whole Google ladder vendor-derived. Google states the multipliers but attaches NO PERIOD to them — no weekly or monthly quota is guaranteed — and its own page says each product has its own limits and that Pro and Ultra members may buy extra AI credits when they hit them. A ratio with no period is not a ceiling this board can rank on, so the ladder is modelled proportional to price. Pro's own position is unchanged: it is the anchor.

  8. 8

    Google AI Ultra (5×)

    $100/mo · Gemini 3 Pro

    ~9.5×

    Google AI Ultra (5×)

    rank #8 · Gemini 3 Pro

    ~9.5×value per $ (headline)
    Maxxed range
    9.5×
    Evidence
    Modelled · n=0
    Binding limit
    weekly + 5h windowRolling 5-hour compute windows plus a longer ceiling, with each tier published as a usage RATIO of AI Pro rather than as tokens.
    Maxxed ceiling
    modelled (no quota published)The vendor publishes a usage RATIO between tiers but no absolute quota, so there is still no cycle to saturate (derivation 4). The ratio sets where this tier sits relative to its siblings; it does not say how much either of them is. Same upper-bound caveat as every modelled row.
    Plan terms
    High
    Throughput
    Low · medium volatility
    List price
    $100/mo
    Token value
    $952–$952
    Tied tier
    unmeasured; tied by assumptionit prints the same rate as its other Google tiers because neither is measured and the model scales the dearer one proportional to price
    Sources
    [1][2][3]
    Last checked
    checked Sep 6

    REPLACES the obsolete “Google AI Ultra $249.99” row, which described a product structure Google retired at I/O on 19 May 2026. Ultra is now two tiers, $100 and $200. Google's support page says Ultra gives “5x or 20x usage quota in Gemini and Google Antigravity compared to the Google AI Pro plan” with no period attached, so this board does not rank on it (see the 20× row). This tier is unaffected either way: $100 is 5.0× Pro's price, so price-proportional modelling and the stated 5× give the same number, and the row lands on Pro's value per dollar — headroom, not a better rate.

  9. 9

    Google AI Ultra (20×)

    $200/mo · Gemini 3 Pro

    ~9.5×

    Google AI Ultra (20×)

    rank #9 · Gemini 3 Pro

    ~9.5×value per $ (headline)
    Maxxed range
    9.5×
    Evidence
    Modelled · n=0
    Binding limit
    weekly + 5h windowRolling 5-hour compute windows plus a longer ceiling, with each tier published as a usage RATIO of AI Pro rather than as tokens.
    Maxxed ceiling
    modelled (no quota published)The vendor publishes a usage RATIO between tiers but no absolute quota, so there is still no cycle to saturate (derivation 4). The ratio sets where this tier sits relative to its siblings; it does not say how much either of them is. Same upper-bound caveat as every modelled row.
    Plan terms
    High
    Throughput
    Low · medium volatility
    List price
    $200/mo
    Token value
    $1,904–$1,904
    Tied tier
    unmeasured; tied by assumptionit prints the same rate as its other Google tiers because neither is measured and the model scales the dearer one proportional to price
    Sources
    [1][2][3]
    Last checked
    checked Sep 6

    REPLACES the obsolete “Google AI Ultra $249.99” row. REVISED 2026-09-06 (lodar): this row used to rank on Google's stated 20× and printed about TWICE Pro's value per dollar — the largest published-ratio-beats-price-ratio gap on the board. Google's actual wording is “Based on your specific Google AI Ultra plan, you get 5x or 20x usage quota in Gemini and Google Antigravity compared to the Google AI Pro plan”: a comparative multiplier with NO PERIOD attached, no guaranteed weekly or monthly quota, on a page that also says each product has its own limits and that members can buy extra AI credits once they hit them. That is not a ceiling, so it cannot carry a rank. The row is now modelled proportional to price like every plan with no published ceiling — $200 is 10× Pro's $19.99 — so it ties with Pro and Ultra 5× on value per dollar. If Google ever states the 20× as a per-cycle quota, or we meter a cycle on it, the upside comes back with evidence behind it.

  10. 10

    GLM Coding Plan (Max)

    $168/mo · GLM-5.3

    ~6.7×

    GLM Coding Plan (Max)

    rank #10 · GLM-5.3

    ~6.7×value per $ (headline)
    Maxxed range
    6.7–13× · ~9.6× at promoHeadline is the saturated end. ~13× is the optimizer's upper bound, not a default.
    Evidence
    Vendor-derived · n=0
    Binding limit
    weekly + 5h windowPrompts are capped per rolling 5-hour window and credits are pooled weekly; whichever empties first binds. Off-peak hours spend credits at 50%.
    Maxxed ceiling
    published creditsDerivation 2, partial. Z.ai publishes a recurring 50% off-peak credit rate, so the same weekly pool buys twice the tokens when autonomous work is scheduled into that window. The band runs from no off-peak scheduling, which is what this row LEADS with because it is what happens without anyone arranging it, up to fully off-peak, which is an upper bound almost nobody reaches. NOT applied: the vendor’s per-request credit deduction coefficients, which it references but does not print — until those are published the pool itself cannot be converted to tokens, so this row still leans on the modelled ceiling for its base.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $168/mo · $118 promo
    Token value
    $1,126–$2,252
    Sources
    [1][2]
    Last checked
    checked Sep 5

    $168/mo list ($117.60 on annual prepay), 28,000 credits/5h + 140,000/week published, ~14× Lite; modelled at that published ratio. Ranked on the LIST monthly price like every other provider on this board, with the discount shown as a badge — we do not rank one provider at a discount and the rest at list. Off-peak deducts 50%, so real headroom is larger than modelled.

  11. 11

    GLM Coding Plan (Pro)

    $80/mo · GLM-5.3

    ~6.0×

    GLM Coding Plan (Pro)

    rank #11 · GLM-5.3

    ~6.0×value per $ (headline)
    Maxxed range
    6.0–12× · ~8.6× at promoHeadline is the saturated end. ~12× is the optimizer's upper bound, not a default.
    Evidence
    Vendor-derived · n=0
    Binding limit
    weekly + 5h windowPrompts are capped per rolling 5-hour window and credits are pooled weekly; whichever empties first binds. Off-peak hours spend credits at 50%.
    Maxxed ceiling
    published creditsDerivation 2, partial. Z.ai publishes a recurring 50% off-peak credit rate, so the same weekly pool buys twice the tokens when autonomous work is scheduled into that window. The band runs from no off-peak scheduling, which is what this row LEADS with because it is what happens without anyone arranging it, up to fully off-peak, which is an upper bound almost nobody reaches. NOT applied: the vendor’s per-request credit deduction coefficients, which it references but does not print — until those are published the pool itself cannot be converted to tokens, so this row still leans on the modelled ceiling for its base.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $80/mo · $56 promo
    Token value
    $483–$965
    Sources
    [1][2]
    Last checked
    checked Sep 5

    $80/mo list ($56 on annual prepay) — corrected from the $79 previously carried here, now read off the vendor pricing table. 12,000 credits/5h + 60,000/week published, ~6× Lite; modelled at that published ratio. Off-peak deducts 50%, so real headroom is larger.

  12. 12

    SuperGrok

    $30/mo · Grok 4.6

    ~4.6×

    SuperGrok

    rank #12 · Grok 4.6

    ~4.6×value per $ (headline)
    Maxxed range
    4.6×
    Community-metered
    ~17× · $502/mo API-equivalentlogged by a third party, published beside the modelled headline and not in place of it — it does not enter the rank
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Binding limit
    weekly poolOne WEEKLY allowance shared across the plan's models and drawn down at a different WEIGHT per model, so the pool is what binds, not any per-model message count — not “unpublished”, which is what this row said before. ATTESTED, NOT READ: x.ai returns 403 to an automated read, so the shape is on the operator's own account rather than off a vendor page we can cite, and the pool SIZE is still unpublished — which is why throughput stays modelled.
    Maxxed ceiling
    modelled (no quota published)The weekly pool's SIZE is not published, only its shape, so there is no number to saturate and nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Low
    Throughput
    Low · high volatility
    List price
    $30/mo
    Token value
    $137–$137
    Tied tier
    same rate as its other xAI tiers
    Last checked
    checked Sep 6

    $30/mo, opaque weekly pool since ~mid-2026 and no published cap of any kind. The old single 1.4× score was flatly contradicted by real agentic usage logs — this row now spans the full floor-to-agentic-ceiling band instead of asserting one number. A community-metered week (see the observation above) lands ~3.6× above even that ceiling; it is published beside the headline rather than folded into it, because one undated third-party log of a saturated week is not the same evidence as the shared model every other row is ranked on. Terms graded low: x.ai returns 403 to automated reads, so the price is cross-checked against secondary trackers and the pool SIZE is unpublished.

  13. 13

    SuperGrok Heavy

    $300/mo · Grok 4.6

    ~4.6×

    SuperGrok Heavy

    rank #13 · Grok 4.6

    ~4.6×value per $ (headline)
    Maxxed range
    4.6×
    Evidence
    Modelled · n=0
    Binding limit
    weekly poolOne WEEKLY allowance shared across the plan's models and drawn down at a different WEIGHT per model, so the pool is what binds, not any per-model message count — not “unpublished”, which is what this row said before. ATTESTED, NOT READ: x.ai returns 403 to an automated read, so the shape is on the operator's own account rather than off a vendor page we can cite, and the pool SIZE is still unpublished — which is why throughput stays modelled.
    Maxxed ceiling
    modelled (no quota published)The weekly pool's SIZE is not published, only its shape, so there is no number to saturate and nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Low
    Throughput
    Low · high volatility
    List price
    $300/mo
    Token value
    $1,366–$1,366
    Tied tier
    same rate as its other xAI tiers
    Sources
    [1][2][3]
    Last checked
    checked Sep 6

    $300/mo, the only consumer tier with the multi-agent Heavy mode and the maximum rate limits. xAI publishes no usage multiple, so modelled proportional to price (10× SuperGrok). Terms graded low: x.ai returns 403 to automated reads.

  14. 14

    GLM Coding Plan (Lite)

    $18/mo · GLM-5.3

    ~4.5×

    GLM Coding Plan (Lite)

    rank #14 · GLM-5.3

    ~4.5×value per $ (headline)
    Maxxed range
    4.5–8.9× · ~6.4× at promoHeadline is the saturated end. ~8.9× is the optimizer's upper bound, not a default.
    Evidence
    Vendor-derived · n=0
    Binding limit
    weekly + 5h windowPrompts are capped per rolling 5-hour window and credits are pooled weekly; whichever empties first binds. Off-peak hours spend credits at 50%.
    Maxxed ceiling
    published creditsDerivation 2, partial. Z.ai publishes a recurring 50% off-peak credit rate, so the same weekly pool buys twice the tokens when autonomous work is scheduled into that window. The band runs from no off-peak scheduling, which is what this row LEADS with because it is what happens without anyone arranging it, up to fully off-peak, which is an upper bound almost nobody reaches. NOT applied: the vendor’s per-request credit deduction coefficients, which it references but does not print — until those are published the pool itself cannot be converted to tokens, so this row still leans on the modelled ceiling for its base.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $18/mo · $12.6 promo
    Token value
    $80.45–$161
    Sources
    [1][2]
    Last checked
    checked Sep 5

    $18/mo list ($12.60 on annual prepay), 2,000 credits/5h + 10,000/week published. Entry tier — the 1× reference for GLM. GLM's cheap retail API rate is why the multiple stays low despite a generous quota: this metric measures retail subsidy, and there is little retail price to subsidise.

  15. 15

    SuperGrok Plus

    $100/mo · Grok 4.6

    ~4.5×

    SuperGrok Plus

    rank #15 · Grok 4.6

    ~4.5×value per $ (headline)
    Maxxed range
    4.5×
    Evidence
    Modelled · n=0
    Binding limit
    weekly poolOne WEEKLY allowance shared across the plan's models and drawn down at a different WEIGHT per model, so the pool is what binds, not any per-model message count — not “unpublished”, which is what this row said before. ATTESTED, NOT READ: x.ai returns 403 to an automated read, so the shape is on the operator's own account rather than off a vendor page we can cite, and the pool SIZE is still unpublished — which is why throughput stays modelled.
    Maxxed ceiling
    modelled (no quota published)The weekly pool's SIZE is not published, only its shape, so there is no number to saturate and nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Low
    Throughput
    Low · high volatility
    List price
    $100/mo
    Token value
    $455–$455
    Sources
    [1][2][3]
    Last checked
    checked Sep 6

    $100/mo, for users actually hitting SuperGrok's limits, plus priority access and 1080p video. xAI publishes no usage multiple, so modelled proportional to price (3.33× SuperGrok). Terms graded low: x.ai returns 403 to automated reads.

  16. 16

    Qwen Token Plan Pro

    $80/mo · Qwen3.8-Max + more

    ~3.9×

    Qwen Token Plan Pro

    rank #16 · Qwen3.8-Max + more

    ~3.9×value per $ (headline)
    Maxxed range
    3.9× · ~4.6× at promo
    Derived
    ~3.9× · $313/mo API-equivalentscaled from a sibling tier's measurement, not measured here · sample ended: the subscription behind it has ended, so it cannot be refreshed. The plan is still on sale; our sample is what stopped.
    Evidence
    Vendor-derived (sample ended) · n=0
    Binding limit
    weekly creditsA 7-day CREDIT pool that resets weekly; the tier ladder is a published credit ratio, not a token figure.
    Maxxed ceiling
    published creditsDerivation 2: the Standard tier’s saturated cycle stretched over the PUBLISHED credit ratio (40,000 vs 10,000 credits per 7 days). The ratio is the vendor’s; the saturation is Standard’s. Pro itself was never metered, so this is vendor-derived, not observed.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $80/mo · $68 promo
    Token value
    $313–$313
    Sources
    [1][2]
    Last checked
    checked Sep 6

    RENAMED and REPRICED 2026-09-06: published as “Qwen Coding Plan (Pro)” at $68, which is the Token Plan Personal Pro tier at its promotional price. List is $80/mo, promo $68, and the board ranks at list. 40,000 credits per 7 days. Scaled at 4× Standard because that is the published CREDIT ratio (40,000 vs 10,000) — not the price ratio, which is 3.2× at list, so Pro buys about 25% more value per dollar than Standard. Valued at qwen3.8-max's published API rate ($2/M in, $6/M out, $0.25/M cache read).

  17. 17

    Qwen Token Plan Standard

    $25/mo · Qwen3.8-Max + more

    ~3.1×

    Qwen Token Plan Standard

    rank #17 · Qwen3.8-Max + more

    ~3.1×value per $ (headline)
    Maxxed range
    3.1× · ~4.3× at promo
    5dive measurement
    ~3.1× · $78.27/mo API-equivalentbeside the modelled headline, not in place of it · Sep 6 · sample ended: the subscription behind it has ended, so it cannot be refreshed. The plan is still on sale; our sample is what stopped.
    Evidence
    Observed (sample ended) · n=1 · seen Sep 6
    Binding limit
    weekly creditsA 7-day CREDIT pool that resets weekly; the tier ladder is a published credit ratio, not a token figure.
    Maxxed ceiling
    metered cycleDerivation 1, the only one on this board built on a real saturation anchor. A metered day drew 8,600,110 tokens and the vendor’s own gauge put it at 22% of the 7-day credit pool, so a fully spent week is ~39.1M tokens and a fully spent month ~170M — 1.61x the modelled agentic ceiling. Published in DOLLARS ($18/week at Qwen’s own rates), because the measured cycle carries its own token mix and must not be re-valued through the board’s modelled one. FROZEN: the subscription ended 2026-09-06.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $25/mo · $18 promo
    Token value
    $78.27–$78.27
    Sources
    [1][2][3]
    Last checked
    checked Sep 6

    RENAMED 2026-09-06: this row and its Pro sibling were published as “Qwen Coding Plan (Standard/Pro)”. Those $18/$68 terms are the Qwen TOKEN Plan Personal, a different product with a different billing method — the Coding Plan is a single $50/mo request-metered plan with no tiers, and it is carried unranked below until we measure it. PRICE CORRECTED: $25/mo list, $18 on the current promotion. The board ranks every provider at LIST price, and this row was ranked at the promo price, which broke its own methodology. 10,000 credits per 7 days is the published ceiling this row is scaled from. It is also the only row with an OBSERVED measurement behind it — see Observed above — and the 91.9% cache-hit rate measured the same day is where the board's agentic cache assumption comes from, so this row is the source of the ceiling assumption every other row now uses. Multi-model plan led by qwen3.8-max alongside qwen3.7-plus, qwen3.6/3.5-plus, qwen3-coder-next/plus and third-party kimi-k2.5 / glm-5 / MiniMax-M2.5.

  18. 18

    Kimi Moderato

    $19/mo · Kimi K3

    ~2.6×

    Kimi Moderato

    rank #18 · Kimi K3

    ~2.6×value per $ (headline)
    Maxxed range
    2.6×
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Binding limit
    monthly poolOne MONTHLY allowance pooled across chat, the console and the coding CLI, so a heavy coding month is spent out of the same pool everything else draws on.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $19/mo
    Token value
    $49.19–$49.19
    Tied tier
    same rate as its other Moonshot AI tiers
    Sources
    [1][2]
    Last checked
    checked Sep 6

    $19/mo on the INTERNATIONAL kimi.ai catalog. CORRECTION 2026-09-06: this row previously carried ¥99 ≈ $13.94 and said the $19 figure was a currency misreading. That was wrong. Moonshot runs two official catalogs — international USD on kimi.ai and China CNY on kimi.com — and the board now prices every Kimi row from the international one, because every other plan here is priced from its internationally marketed USD page. Kimi publishes a ~60 agent-use/mo ceiling drawn from a SHARED monthly credit pool, plus separate Kimi Code 5-hour/weekly limits — and community telemetry shows that pool exhausting fast under sustained use (one heavy user reported ~91% gone in 9 days). So while K3's API is flagship-priced, the plan cannot SUSTAIN the generic agentic throughput an unmetered plan can: sustained value is haircut to ~0.2× the sustained agent-load model. Community-observed: the haircut rests on third-party logs, not on our own meter, which is why the throughput confidence is low. DIVE-4028: this row used to carry TWO haircuts, a gentler one at the retired human-chat floor and this one at the agentic end. With one workload on the board only the sustained-agent figure survives, so the row prints a point estimate and leans on its throughput confidence rather than on a band that spanned two different readers.

  19. 19

    Kimi Vivace

    $199/mo · Kimi K3

    ~2.6×

    Kimi Vivace

    rank #19 · Kimi K3

    ~2.6×value per $ (headline)
    Maxxed range
    2.6×
    Evidence
    Modelled · n=0
    Binding limit
    monthly poolOne MONTHLY allowance pooled across chat, the console and the coding CLI, so a heavy coding month is spent out of the same pool everything else draws on.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $199/mo
    Token value
    $514–$514
    Tied tier
    same rate as its other Moonshot AI tiers
    Sources
    [1]
    Last checked
    checked Sep 6

    NEW row — $199/mo, the top tier of the INTERNATIONAL kimi.ai catalog. It was previously left off the board on the mistaken belief that it did not exist; it does, internationally, and it has no China-catalog equivalent. Moonshot publishes no pool figure for it, so it is modelled proportional to price against Moderato (10.5× the price, so 10.5× Moderato's already-haircut scale), which by construction prints Moderato's value per dollar: the price buys headroom, not a better rate. Low throughput confidence.

  20. 20

    Qwen Token Plan Lite

    $8/mo · Qwen3.8-Max + more

    ~2.4×

    Qwen Token Plan Lite

    rank #20 · Qwen3.8-Max + more

    ~2.4×value per $ (headline)
    Maxxed range
    2.4× · ~3.3× at promo
    Evidence
    Vendor-derived · n=0
    Binding limit
    weekly creditsA 7-day CREDIT pool that resets weekly; the tier ladder is a published credit ratio, not a token figure.
    Maxxed ceiling
    published creditsDerivation 2: Standard’s saturated cycle at the published Lite credit ratio (a quarter of Standard’s pool). Vendor ratio, borrowed saturation — Lite was never metered.
    Plan terms
    High
    Throughput
    Med · medium volatility
    List price
    $8/mo · $6 promo
    Token value
    $19.57–$19.57
    Sources
    [1]
    Last checked
    checked Sep 6

    NEW row — $8/mo list ($6 promo), 2,500 credits per 7 days. Scaled at 0.25× Standard on the published CREDIT ratio (2,500 vs 10,000). That is worse than its price ratio (8/25 = 0.32× Standard's price for 0.25× the credits), so Lite is the WEAKEST value per dollar of the three Qwen tiers, not the best: the cheap entry price buys proportionally fewer credits. Cheapest way in, not cheapest way to buy tokens.

  21. 21

    Kimi Allegro

    $99/mo · Kimi K3

    ~2.2×

    Kimi Allegro

    rank #21 · Kimi K3

    ~2.2×value per $ (headline)
    Maxxed range
    2.2×
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Binding limit
    monthly poolOne MONTHLY allowance pooled across chat, the console and the coding CLI, so a heavy coding month is spent out of the same pool everything else draws on.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $99/mo
    Token value
    $221–$221
    Sources
    [1][2]
    Last checked
    checked Sep 6

    $99/mo on the international kimi.ai catalog (was carried at ¥699 ≈ $98.45 from the China catalog — close in dollars, but a different catalog and a different tier ladder). ~360 agent-uses/mo (~6× Moderato's pool), yet a heavy user documented ~91% of the monthly pool gone in 9 days WITHOUT hitting the 5h or weekly caps — the larger pool does not convert into sustained agentic value. Haircut for the pool cap: ~0.9× the sustained agent-load model.

  22. 22

    Kimi Allegretto

    $39/mo · Kimi K3

    ~2.1×

    Kimi Allegretto

    rank #22 · Kimi K3

    ~2.1×value per $ (headline)
    Maxxed range
    2.1×
    Evidence
    Community-observed · n=1 · date not establishedthe third-party log behind this row is undated, so we cannot say how old it is — treat the band, not the date, as what it supports
    Binding limit
    monthly poolOne MONTHLY allowance pooled across chat, the console and the coding CLI, so a heavy coding month is spent out of the same pool everything else draws on.
    Maxxed ceiling
    modelled (no quota published)No published quota to saturate, so nothing better than the workload model exists (derivation 4). The figure here is the modelled sustained-agent ceiling: an UPPER BOUND on a modelled token mix, not a measurement of saturation. Read it as “we cannot see this plan’s ceiling”, not as “this plan has no headroom”.
    Plan terms
    Med
    Throughput
    Low · high volatility
    List price
    $39/mo
    Token value
    $83.62–$83.62
    Sources
    [1][2]
    Last checked
    checked Sep 6

    $39/mo on the international kimi.ai catalog (was carried at ¥199 ≈ $28.03 from the China catalog). Same SHARED monthly credit pool + Kimi Code 5h/weekly limits as Moderato with a larger allowance (~2× the pool), but community reports show the higher Kimi tiers exhausting the pool too, so the extra allowance does not scale cleanly into sustained agentic value. Haircut like Moderato for the pool cap: ~0.34× the sustained agent-load model.

Glass-box methodology

One formula. Every number sourced or flagged.

Value multiple = the API-equivalent retail value of a month's modelled usage, divided by the plan's monthly list price.

Why subsidised plans win it

Value per dollar is really subsidy per dollar. Plans riding an expensive API top the list.

Every ceiling is modelled

No plan publishes a token quota, so usage is modelled from one stated workload: a coding agent running continuously until the plan's own rate windows stop it.

What API-equivalent means

The public retail API price of the underlying model, not the provider's own inference cost — retail subsidy, not margin. No quality reweighting: a weaker model's tokens count the same as a flagship's.

Which is why the same month of use is worth far more at an expensive-API flagship (Claude Opus, ChatGPT Sol) than at a cheap one (GLM, Qwen), and why plans on pricey models — plus Kimi, a flagship-priced API sold cheap — top the list.

Why most of these are estimates

Almost no provider publishes a token quota, so we model a month of an agent running continuously until the plan's own limits stop it. That model is an upper bound, not a measurement.

Where we have metered a plan end to end, the row publishes measured dollars and is marked as measured. A metered row can therefore read LOWER than an estimated one: measurement is a correction, not a bonus.

How the floor and ceiling are modelled

No consumer plan publishes a token quota, so every plan's monthly usage is MODELLED from one stated, reproducible workload: SUSTAINED AGENT LOAD — a coding agent running continuously at ~40,000 tokens/turn (35K input + 5K output), ~30 turns/hour, until the plan's own rate windows stop it ≈ 105.6M tokens/mo on an entry plan. Each row scales that by its own sustained ceiling relative to its provider's entry tier: a published request or credit ceiling where one exists, otherwise proportional to PRICE — never to an advertised “N× usage” session-burst label. Input is split at the provider's PUBLISHED cache-read rate, because a measured 91.9% of an agentic workload's input is re-sent context billing as a cache read, typically a tenth of fresh input. What is left between floor and ceiling is uncertainty in that plan's own ceiling, not a change of workload. Agents are bounded by the QUOTA, not by the clock: on the one plan we could check, a day of round-the-clock work drew 22% of a SEVEN-day allowance, so saturating the quota is worth ~1.61x this modelled volume, not six-fold. That lift is applied only where a published quota or our own meter anchors it. Each row LEADS with the QUOTA-SATURATED end of its own band, because that is where an agent that simply runs ends up; the full range, the confidence and the check date sit behind that row’s (i). Every row declares where its throughput input came from — observed on our own meter, vendor-derived, third-party logged, or modelled — plus its sample count and, where the source carries a date, when it was last observed. Weekly quotas convert at 4.348 weeks/mo (365.2425/12/7), not 4. Where a provider meters chat, its coding agent and its work product on SEPARATE allowances, the row names the ONE surface it ranks — the provider's coding surface — and never blends two allowances. Where two sibling tiers print the same multiple only because the model divided price by price, the row says “unmeasured; tied by assumption”, not “same rate”.

What maxxed means, and why a metered row can read lower

Every row's band runs from the plan's binding quota saturated by SUSTAINED AGENT LOAD — a coding agent running continuously at 40,000 tokens a turn, 30 turns an hour, hitting the rate windows naturally — up to that same load PLUS what an optimizer adds: schedulable work moved into a RECURRING off-peak window, prompt caches preserved, usage valued at the highest API-priced model the plan includes. The row LEADS with the SATURATED end: an agent reaches it by default, while almost nobody runs the optimizer's tactics, so a mean would flatter every row. Temporary promotions and one-off bonus credits are excluded everywhere — a permanent score cannot rest on terms that expire. Each row says which quota binds it first and how its ceiling was obtained, strongest first: (1) a cycle we metered against the vendor's own gauge; (2) a published credit pool converted through the vendor's deduction coefficients; (3) a published request cap times measured tokens per request; (4) nothing better exists, so the modelled ceiling stands in and the row says so. Class 4 is an upper bound on a modelled token mix, not a saturation measurement. Where a row WAS metered it publishes the measured DOLLARS and never re-values the tokens: a real cycle carries its own mix, and this board's modelled mix assumes far more output than agent traffic produces. On the one row with both readings the gap is 3.15x, almost entirely because the model assumes 12.5% of tokens are output where agent traffic runs under 1%. So read a MODELLED row as an upper bound hot by roughly that factor — correcting the mix on every row is a separate change this release does not make. That is also why a metered row can read LOWER: measurement is a correction, not a bonus.

Where the headline number comes from

The quota-saturated end of that row's range: what an agent that simply runs reaches. The other end adds recurring off-peak scheduling, cache preservation and valuing usage at the priciest model the plan includes — almost nobody does any of that, so a centre between the two would flatter every row. It is still printed behind the row icon as the upper bound.

Where the range is uncertainty about how much a TIER delivers rather than optimizer effort, both ends are already saturated, so that row leads with their geometric mean instead — geometric because these are multipliers. The row says which it did. Either way the headline reads no input the range does not, which is why it prints with a tilde.

Why some tiers tie

Where a provider publishes no ceiling, each of its tiers is modelled proportional to price — and price divides straight back out, so those tiers print the same headline by construction. Not a rounding artefact: the dearer ChatGPT and SuperGrok tiers buy headroom, not a better rate. Tied rows are marked and listed cheapest first. Providers that do publish a ceiling — Kimi, GLM, Qwen — separate on their own.

Which price the value is divided by

The list monthly price, for every provider, so no two rows are ever compared at different discount states. Where a provider is currently advertising a lower rate, that price and the headline it would produce both appear in the row's details. They do not move the rank.

How to read a wide band, and what moves the order

The band is throughput uncertainty in that plan's own sustained ceiling — both ends describe the same reader, a continuously running agent. A wide band is not a weak rank: a row whose ends move apart (Claude Max, Google AI Ultra) is one where the ceiling itself is contested.

What does move the order is each row's scale factor. Where a provider publishes no ceiling we scale proportional to price, never to an advertised N× usage label: those are five-hour-session bursts that weekly caps stop you sustaining across a month, which is why an entry plan can out-value its own higher tier.

What these figures are not

Quotas are dynamic. Every figure here is a SNAPSHOT of what a plan's published terms implied on the date in its Last-checked column, not an entitlement you can hold a provider to. These plans re-price and re-cap without notice, sometimes weekly, and none of them publish a token quota at all.

This board prices TOKENS, not experience. Latency, queueing, 429/rate-limit behaviour, uptime and model quality are all excluded from the multiple — a plan can rank high here and still be unpleasant to work on, and model strength is a separate board.

Every row says where its throughput input came from. Observed means we metered it on our own subscription against the provider's quota meter; where that subscription has ended the row says sample ended and carries the date, so a frozen sample never reads as a live one. Vendor-derived means the vendor publishes a credit or request ceiling, or a usage ratio, and we applied it. Community-observed rests on third-party logs, which widens the band and caps the confidence. Modelled is the fallback: no telemetry, no published ceiling, so the tier is scaled proportional to price.

An observation is printed beside the modelled headline, never in place of it — the ranking stays on the one formula every row shares. Each row also carries its sample count, the date of the last observation where its source is dated, how fast that plan's terms move, and grades plan-terms confidence separately from throughput confidence: a price you can read off a vendor page tells you nothing about the quota behind it.

Kimi is priced here from Moonshot's INTERNATIONAL catalog (kimi.ai, USD), because every other plan on this board is priced from its internationally marketed page. Moonshot also sells a separate China catalog (kimi.com, CNY) with a different tier ladder — Andante CNY 49, Moderato CNY 99, Allegretto CNY 199, Allegro CNY 699 — which at today's rate is materially cheaper and would rank Kimi differently. Which catalog you can buy from depends on your region.

Own the box, max the plan

Point your best plan at a server you own.

Run the subscription you already pay for as an always-on AI teammate.