AI Benchmark

November 2026 · run 2026-11-10

Best value this month

Claude

At everyday quality, every dollar on a Claude plan finishes 24.1 tasks a month, 1.32× ChatGPT. Run Opus 5.5 at medium. Pick the tier by how much you use, because no tier is cheaper per task than another.

Mostly chat, some code$20

Claude Pro

Sonnet 5.5 · medium

1,590 tasks/mo

1.20× ChatGPT Plus (Sol 6.1 low)

Agent coding most days$100

Claude Max

Opus 5.5 · medium

2,341 tasks/mo

1.32× ChatGPT Pro 100 (Sol 6.1 high)

If OpenAI's announced 2× is real: ChatGPT Pro 100 3,537. Not metered yet.

Agent coding all day$200

Claude Ultra

Opus 5.5 · medium

4,784 tasks/mo

1.35× ChatGPT Pro 200 (Sol 6.1 high)

If OpenAI's announced 2× is real: ChatGPT Pro 200 7,074. Not metered yet.

Hardest problems, top quality$200

Claude Ultra

Opus 5.5 · xhigh

1,939 tasks/mo

4.23× ChatGPT Pro 200 (Astra max)

API dollars are not work done

Each lab prices its own tokens. Divide by what a finished task costs and the gap changes size.

API dollars a monthwhat the tokens would cost on the API
$20Claude Pro · Opus 5.5: $1,181$1,181ChatGPT Plus · Sol 6.1: $213$213$100Claude Max · Opus 5.5: $5,739$5,739ChatGPT Pro 100 · Sol 6.1: $1,045$1,045$200Claude Ultra · Opus 5.5: $11,726$11,726ChatGPT Pro 200 · Sol 6.1: $2,090$2,090
Tasks a month at everyday agent coding qualityindex ≥ 48.2 (Opus 5.5 medium), best model in each plan
$20Claude Pro · Opus 5.5 medium: 482482ChatGPT Plus · Sol 6.1 high: 361361$100Claude Max · Opus 5.5 medium: 2,3412,341ChatGPT Pro 100 · Sol 6.1 high: 1,7691,769$200Claude Ultra · Opus 5.5 medium: 4,7844,784ChatGPT Pro 200 · Sol 6.1 high: 3,5373,537
ClaudeChatGPT

The effort dial moves more than the brand

Tasks a month on the $200 plans as you turn effort up. Hover a point for score and cost per task.

05,00010,00015,000lowmediumhighxhighmaxOpus 5.5 @ low: 12,020 tasks/mo · index 34.8 · $0.98/taskOpus 5.5 @ medium: 4,784 tasks/mo · index 48.2 · $2.45/taskOpus 5.5 @ high: 3,384 tasks/mo · index 51.8 · $3.46/taskOpus 5.5 @ xhigh: 1,939 tasks/mo · index 57.4 · $6.05/taskOpus 5.5 @ max: 1,168 tasks/mo · index 59.1 · $10.04/taskOpus 5.5Sol 6.1 @ low: 12,944 tasks/mo · index 34.8 · $0.16/taskSol 6.1 @ medium: 5,817 tasks/mo · index 46.3 · $0.36/taskSol 6.1 @ high: 3,537 tasks/mo · index 50.3 · $0.59/taskSol 6.1 @ xhigh: 2,736 tasks/mo · index 52.5 · $0.76/taskSol 6.1 @ max: 1,643 tasks/mo · index 54.6 · $1.27/taskSol 6.1
Claude Ultra · Opus 5.5ChatGPT Pro 200 · Sol 6.1

Every tier

Tasks counted at everyday agent coding quality (Artificial Analysis index ≥ 48.2) with the best model each plan includes. Faded rows have no meter reading.

Claude Anthropic

TierPrice/moAPI value/moTokens/moTasks/mo at everyday agent codingTasks per $
Claude Pro5-hour and weekly limits$20$1,181Opus 5.52.9B482Opus 5.5 medium24.1
Claude Maxwas Claude Max 5x5-hour and weekly limits; Fable up to half the weekly limit$100$5,739Opus 5.514.0B2,341Opus 5.5 medium23.4
Claude Ultrawas Claude Max 20xWeekly limit only (no 5-hour limit for verified accounts); Fable up to half the weekly limit$200$11,726Opus 5.528.6B4,784Opus 5.5 medium23.9

ChatGPT OpenAI

TierPrice/moAPI value/moTokens/moTasks/mo at everyday agent codingTasks per $
ChatGPT GoLimited Codex access$8No published meter reading · no included model reaches index 48.2Includes Luna
ChatGPT Plus5-hour and weekly limits$20$213Sol 6.11.0B361Sol 6.1 high18.0
ChatGPT Pro 1002× announced 10-29, unmeteredWeekly limit only$100$1,045Sol 6.15.1B1,769Sol 6.1 high17.7
ChatGPT Pro 2002× announced 10-29, unmeteredWeekly limit only$200$2,090Sol 6.110.2B3,537Sol 6.1 high17.7
ChatGPT Pro 5002× announced 10-29, unmeteredWeekly limit only; Ultrafast mode$500$5,390Sol 6.126.3B9,120Sol 6.1 high18.2

Google AI Google

Unmeasured. Nobody has published a usage-meter reading for these plans, so their capacity cannot be ranked. Tiers are listed with the models they include.

TierPrice/moAPI value/moTokens/moTasks/mo at everyday agent codingTasks per $
Google AI PlusNot published as a number$4.99No published meter reading · no included model reaches index 48.2Includes 3.8 Flash
Google AI ProNot published as a number$20No published meter reading · cheapest at the bar: Opus 5.5 medium, $2.45/task at API priceIncludes Argon, Opus 5.5, Sonnet 5.5, 3.8 Flash
Google AI Ultra 5x"5x higher usage limits vs. AI Pro"$100No published meter reading · cheapest at the bar: Opus 5.5 medium, $2.45/task at API priceIncludes Carbon, Argon, Opus 5.5, Sonnet 5.5, 3.8 Flash
Google AI Ultra 20x"20x higher usage limits vs. AI Pro"$200No published meter reading · cheapest at the bar: Opus 5.5 medium, $2.45/task at API priceIncludes Carbon, Argon, Opus 5.5, Sonnet 5.5, 3.8 Flash

Grok xAI

TierPrice/moAPI value/moTokens/moTasks/mo at everyday agent codingTasks per $
X PlusWeekly pool shared by Grok and Cursor$8No published meter reading · no included model reaches index 48.2Includes Grok 4.7
X Premium"4x more usage compared to Plus"$30No published meter reading · no included model reaches index 48.2Includes Grok 4.7
X Super"15x more usage compared to Plus"$100No published meter reading · no included model reaches index 48.2Includes Grok 4.7
X Ultra"50x more usage compared to Plus"$200No published meter reading · no included model reaches index 48.2Includes Grok 4.7
SuperGrok Heavyretired$300$5,832Grok 4.710.3Bno included model reaches the bar

Which model for which job

Anthropic's guide names a starting model per job. Its benchmark score becomes the bar; each plan gets the model and effort that clears it for the most tasks.

JobAnthropic's pick → barClaude UltraChatGPT Pro 200Google AI Ultra 20xX Ultra
Triage & extractionClassify, route, tag, pull fields outHaiku 5.5 mediumindex ≥ 24.6Sonnet 5.5 low24,504/moSol 6.1 low12,944/moSonnet 5.5 low$0.51/task · unmeteredGrok 4.7 low$2.92/task · unmetered
Chat & quick answersInteractive sessions you wait onSonnet 5.5 mediumindex ≥ 30.8Sonnet 5.5 medium16,048/moSol 6.1 low12,944/moSonnet 5.5 medium$0.78/task · unmeteredGrok 4.7 low$2.92/task · unmetered
Scoped codingBug fixes, spec-driven featuresSonnet 5.5 highindex ≥ 41.5CursorBench 47.8% · $1.20/taskSonnet 5.5 high7,658/moSol 6.1 medium5,817/moSonnet 5.5 high$1.64/task · unmeterednothing reaches it
Everyday agent codingMulti-file features, debuggingOpus 5.5 mediumindex ≥ 48.2CursorBench 52.5% · $2.91/taskOpus 5.5 medium4,784/moSol 6.1 high3,537/moOpus 5.5 medium$2.45/task · unmeterednothing reaches it
Hardest problemsMigrations, architecture, researchFable 5.2 mediumindex ≥ 56.4CursorBench 54.8% · $4.66/taskOpus 5.5 xhigh1,939/moAstra max459/moOpus 5.5 xhigh$6.05/task · unmeterednothing reaches it

Method

  1. MeterTokens a month each plan holds, read by SemiAnalysis (2026-10-05) from the providers' own usage meters. Never from a price page.
  2. PriceThose tokens at API list price, blended by an agentic mix: 0.43% input, 96.6% cache reads, 2.63% cache writes, 0.34% output.
  3. TasksAPI dollars ÷ Intelligence Index v5.0.1 cost per task at each effort. CursorBench 4.0 is the coding cross-check.
  4. ComparePlans at the same quality bar, at the same price. Announced but unmetered changes are shown, never counted.

Changed since the meter reading

  • SemiAnalysis detected an Anthropic A/B test that cuts monthly API value by about 20% and raises the 5-hour limit by about 20% on some accounts. source
  • OpenAI announced it doubled all usage limits on Pro 100, 200 and 500. No meter reading after the change has been published, so the measured figures stay; the doubling is shown only as a labelled what-if. source
  • Max 5x renamed Max, Max 20x renamed Ultra, same prices; Ultra loses the 5-hour limit for verified accounts; weekly limits unchanged. Readings carry over to the renamed plans. source
  • Claude Fable 5.2 replaced Fable 5.1 as the flagship. No meter reading for Fable 5.2 has been published, so Fable on Claude plans is unmeasured until one is. source

Checked against the video

The method was rebuilt from $200 for ChatGPT, Claude or Gemini? I Did the Math and run on the inputs the video used. None of its numbers is an input.

45/48

video figures reproduced with the inputs as of 2026-10-26

27/48

still the same with November 2026 inputs

Replay of the video (2026-10-26 inputs), every figure
FigureVideoOursWhy
Claude Max 20x API value (Opus 5.5)$11,726$11,726same
ChatGPT Pro 200 API value (Sol 6.1)$2,084$2,090same
API-value ratio at $2005.60×5.61×same
Claude Pro API value (Opus 5.5)$1,178$1,181same
ChatGPT Plus API value (Sol 6.1)$211$213same
Claude Max 5x API value (Opus 5.5)$5,725$5,739same
ChatGPT Pro 100 API value (Sol 6.1)$1,055$1,045same
Same ratio at every tieryesyessame
Blended Opus 5.5 price per MTok (agentic)$0.40$0.41same
Blended Sol 6.1 price per MTok (agentic)$0.20$0.20same
Max 20x tokens a month (billions)29.0B28.6Bsame
Pro 200 tokens a month (billions)10.0B10.2Bsame
Token ratio at $2002.80×2.80×same
Opus API value per $1 of fee$58.90$59.03same
Sol API value per $1 of fee$10.55$10.66same
No volume discount at either labyesyessame
Max 20x tokens in one 5-hour window (millions)1,740M1,756Msame
Max 20x tokens in one week (millions)6,579M6,584Msame
Full 5-hour windows per week3.80×3.75×same
Matched-score pairclaude-opus-5.5@medium vs gpt-6.1-sol@xhighclaude-opus-5.5@medium vs gpt-6.1-sol@xhighsame
Cost-per-task ratio at matched score3.40×3.40×same
Max 20x tasks a month (Opus medium)8,7508,777same
Pro 200 tasks a month (Sol xhigh)5,3405,321same
Max 20x tasks per working day400399same
Pro 200 tasks per working day240242same
Task ratio at matched score1.60×1.65×same
Opus 5.5 max-effort index score58.057.6same
Opus max vs medium cost per task4.50×4.48×same
Max 20x tasks a month at max effort1,9601,960same
Pro 200 tasks a month at max effort2,8902,887same
At max effort ChatGPT does more tasksyesyessame
Max 20x Fable 5.1 API value$2,485$2,506same
Pro 200 Astra API value$2,897$2,865same
Flagship matched pairclaude-fable-5.1@xhigh vs gpt-6-astra@maxclaude-fable-5.1@xhigh vs gpt-6-astra@maxsame
Max 20x Fable tasks a month416419same
Pro 200 Astra tasks a month889879same
Astra tasks if $2,897 were the old allowance444440sameSemiAnalysis labels 1.9B as the Pro 200 allowance after the Sep 29 cut, so the halving hedge is not needed.
Pro 500 Astra vs old Pro 2001.21×1.21×same
Ultrafast burn rate8.00×8.00×same
Gemini 4 Argon index score53.052.6same
Gemini 4 Argon cost per task$1.99$1.62differentArtificial Analysis changed Gemini 4 Argon's cost per task during Oct 26: the Wayback capture at 2026-10-26 01:02 UTC shows $1.99 (the video's figure), the 21:23 capture this replay uses shows $1.62. https://web.archive.org/web/20261026010237/https://artificialanalysis.ai/leaderboards/models
Argon API price doubles on Dec 1yesyessame
Google AI Ultra has no published meter readingyesyessame
Max 20x Opus API value, chat mix$9,200$9,203same
Max 20x API value on the A/B-test variant$9,385$9,380same
Max 5x tasks a month (Opus medium)4,2704,295same
June: ChatGPT Pro ~$14,000 vs Claude Max 20x ~$8,000not recomputed–n/aNot recomputed: needs SemiAnalysis' June readings and June models, which this run does not fetch.
Opus max burns the weekly allowance ~11x faster than medium11.00×–n/aNot reproducible: Artificial Analysis publishes cost per task, not tokens per task, and cost per task gives 4.5x, not 11x.
This month vs the video, with what moved
FigureVideoOursWhy
Claude Max 20x API value (Opus 5.5)$11,726$11,726same
ChatGPT Pro 200 API value (Sol 6.1)$2,084$2,090same
API-value ratio at $2005.60×5.61×same
Claude Pro API value (Opus 5.5)$1,178$1,181same
ChatGPT Plus API value (Sol 6.1)$211$213same
Claude Max 5x API value (Opus 5.5)$5,725$5,739same
ChatGPT Pro 100 API value (Sol 6.1)$1,055$1,045same
Same ratio at every tieryesyessame
Blended Opus 5.5 price per MTok (agentic)$0.40$0.41same
Blended Sol 6.1 price per MTok (agentic)$0.20$0.20same
Max 20x tokens a month (billions)29.0B28.6Bsame
Pro 200 tokens a month (billions)10.0B10.2Bsame
Token ratio at $2002.80×2.80×same
Opus API value per $1 of fee$58.90$59.03same
Sol API value per $1 of fee$10.55$10.66same
No volume discount at either labyesyessame
Max 20x tokens in one 5-hour window (millions)1,740M1,756Msame
Max 20x tokens in one week (millions)6,579M6,584Msame
Full 5-hour windows per week3.80×3.75×same
Matched-score pairclaude-opus-5.5@medium vs gpt-6.1-sol@xhighclaude-opus-5.5@medium vs gpt-6.1-sol@mediumdifferentscores [51.243, 51.038] → [48.208, 46.347]; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Cost-per-task ratio at matched score3.40×6.82×differentusd_per_task [1.336, 0.3929] → [2.4511, 0.3594]; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Max 20x tasks a month (Opus medium)8,7504,784differentusd_per_task 1.336 → 2.451; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Pro 200 tasks a month (Sol xhigh)5,3405,817differentusd_per_task 0.3929 → 0.3594; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Max 20x tasks per working day400217differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Pro 200 tasks per working day240264differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Task ratio at matched score1.60×0.82×differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Opus 5.5 max-effort index score58.059.1differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Opus max vs medium cost per task4.50×4.09×differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Max 20x tasks a month at max effort1,9601,168differentusd_per_task 5.982 → 10.04; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Pro 200 tasks a month at max effort2,8901,643differentusd_per_task 0.7242 → 1.272; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
At max effort ChatGPT does more tasksyesyessame
Max 20x Fable 5.1 API value$2,485–n/aFable 5.1 left the Claude plans on 2026-11-09 (replaced by Fable 5.2) and is not in this Artificial Analysis index; no meter reading for Fable 5.2 has been published, so the flagship comparison waits for one.
Pro 200 Astra API value$2,897$2,865same
Flagship matched pairclaude-fable-5.1@xhigh vs gpt-6-astra@max–n/aFable 5.1 left the Claude plans on 2026-11-09 (replaced by Fable 5.2) and is not in this Artificial Analysis index; no meter reading for Fable 5.2 has been published, so the flagship comparison waits for one.
Max 20x Fable tasks a month416–n/aFable 5.1 left the Claude plans on 2026-11-09 (replaced by Fable 5.2) and is not in this Artificial Analysis index; no meter reading for Fable 5.2 has been published, so the flagship comparison waits for one.
Pro 200 Astra tasks a month889–n/aFable 5.1 left the Claude plans on 2026-11-09 (replaced by Fable 5.2) and is not in this Artificial Analysis index; no meter reading for Fable 5.2 has been published, so the flagship comparison waits for one.
Astra tasks if $2,897 were the old allowance444–n/aFable 5.1 left the Claude plans on 2026-11-09 (replaced by Fable 5.2) and is not in this Artificial Analysis index; no meter reading for Fable 5.2 has been published, so the flagship comparison waits for one.
Pro 500 Astra vs old Pro 2001.21×1.21×same
Ultrafast burn rate8.00×8.00×same
Gemini 4 Argon index score53.050.4differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Gemini 4 Argon cost per task$1.99$9.95differentArtificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
Argon API price doubles on Dec 1yesyessame
Google AI Ultra has no published meter readingyesyessame
Max 20x Opus API value, chat mix$9,200$9,203same
Max 20x API value on the A/B-test variant$9,385$9,380same
Max 5x tasks a month (Opus medium)4,2702,341differentusd_per_task 1.336 → 2.451; Artificial Analysis re-indexed (Intelligence Index v4.3.2 → Intelligence Index v5.0.1)
June: ChatGPT Pro ~$14,000 vs Claude Max 20x ~$8,000not recomputed–n/aNot recomputed: needs SemiAnalysis' June readings and June models, which this run does not fetch.
Opus max burns the weekly allowance ~11x faster than medium11.00×–n/aNot reproducible: Artificial Analysis publishes cost per task, not tokens per task, and cost per task gives 4.5x, not 11x.

Sources

Runs