Models · Anthropic · Out sinceReleased 22 Sep 2026

Claude Opus 5.5#2 for writing.#2 for writing, best at high effort.

Available on OpenRouterOn OpenRouterAvailable nowGenerally availableClosed (can't be downloaded)ProprietaryOur pick: The best writer you can use todayPick: Best for writing & creativityOur pick: The best analyst for long, demanding documentsPick: Best for research & analysisOur pick: The best model for coding right nowPick: Best for coding right now

Where you can use it:In apps:Claude (Team plan) · Opus 5.5

Claude Opus 5.5 is made by Anthropic. Among the models we track it ranks #2 for writing, #2 for coding, #4 for research and analysis. It's expensive to use. In blind tests where thousands of people compare two anonymous texts, people prefer Claude Opus 5.5's writing over every other model you can use today. It's also the strongest at long, well-built stories and texts.

API id claude-opus-5-5. Default effort = medium (unlike other Claude models, which default to high). Adaptive thinking always on (thinking cannot be disabled). Cache reads $0.20/M, cache writes $5/M. Fast mode (up to 2.5x speed) $8/$40. Cyber/bio safeguards with fallback to Opus 4.8 / Opus 5. Knowledge cutoff Jun 2026.

Writing & creativity: 86.6 / 100 · #2
Research & analysis: 81.2 / 100 · #4
Coding: 89.9 / 100 · #2
Price: Expensive$4 / $20
Price per 1M (blended)Blended / 1M: $8
MemoryContext: 1M
Longest answerMax output: 128K
Test resultsResults: 168 (110 independent110 indep.)
Out sinceReleased: 22 Sep 2026
Made byVendor: Anthropic
UnderstandsInputs: text, image

Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.

Thinking levels

Reasoning settings

LowMediumHighExtra highMax

Highlighted: where it did best for writing (High thinking).

Highlighted: dominant setting in its writing composite.

Links

Using it through OpenRouter

On OpenRouter

OpenRouter is a service that gives access to many AI models in one place. This model is listed there as anthropic/claude-opus-5.5.

Route it as anthropic/claude-opus-5.5 at $4 in / $20 out per 1M tokens, 1M context. Listed since 22 Sep 2026.

include_reasoningmax_completion_tokensmax_tokensreasoningreasoning_effortresponse_formatstopstructured_outputstemperaturetool_choicetoolsverbosity

Thinking level

Reasoning effort

How long should Claude Opus 5.5Where Claude Opus 5.5think?peaks.

Each chart is one test. Left to right, the model thinks longer; higher is a better result. The yellow ring marks where it did best. Thinking level is how long the AI thinks before it answers. Higher is slower and costs more, and it isn't always better.

Score per reasoning setting on each benchmark with ≥2 settings; ring = peak. Canonical results (independent over vendor).

Terminal-Bench 4.0

Best at MaxBest at Max

CursorBench

Best at MaxBest at Max

FrontierCode

Best at Medium, worse abovePeaks at Medium

FrontierCode v1.1 (Extended)

Best at Medium, worse abovePeaks at Medium

SciCode

Best at MaxBest at Max

SWE-bench Pro (Anthropic internal subset)

Best at HighBest at High

Artificial Analysis Intelligence Index

Best at MaxBest at Max

AutomationBench

Best at MaxBest at Max

AA-Briefcase v1.1

Best at MaxBest at Max

AA-LCR

Best at Extra highBest at Extra high

AA-Omniscience Accuracy

Best at MaxBest at Max

AA-Omniscience Hallucination Rate

Best at MaxBest at Max

Lower is better on this test

AA-Omniscience Index

Best at MaxBest at Max

ARC-AGI-2

Best at High, worse abovePeaks at High

GDP.pdf

Best at High, worse abovePeaks at High

GDPval-AA v2.1

Best at MaxBest at Max

Humanity's Last Exam

Best at MaxBest at Max

WANDR

Best at MaxBest at Max

Every test result

Every result

All the numbers,with where they came from.

Every result we've found for this model, with who measured it and a link to where we read it.

All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.

All results for this model
						NotesNotes
AA-Briefcase v1.1	1285	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	AA-Briefcase v1.1 (Artificial Analysis, long-horizon knowledge work, Elo); vendor-reported cost/task $1.15
AA-Briefcase v1.1	1642	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	AA-Briefcase v1.1 (Artificial Analysis, long-horizon knowledge work, Elo); vendor-reported cost/task $4.4
AA-Briefcase v1.1	1705	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Briefcase v1.1	1705	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	AA-Briefcase v1.1 (Artificial Analysis, long-horizon knowledge work, Elo); vendor-reported cost/task $6.27
AA-Briefcase v1.1	1780	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Briefcase v1.1	1780	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	AA-Briefcase v1.1 (Artificial Analysis, long-horizon knowledge work, Elo); vendor-reported cost/task $12.27
AA-Briefcase v1.1	1822	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Briefcase v1.1	1822	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	AA-Briefcase v1.1 (Artificial Analysis, long-horizon knowledge work, Elo); vendor-reported cost/task $21.05
AA-LCR	80.7%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-LCR accuracy, AA-run (long multi-document reasoning)
AA-LCR	84.3%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-LCR accuracy, AA-run (long multi-document reasoning)
AA-LCR	82.7%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-LCR accuracy, AA-run (long multi-document reasoning)
AA-LCR	84.7%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-LCR accuracy, AA-run (long multi-document reasoning)
AA-LCR	84.7%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-LCR accuracy, AA-run (long multi-document reasoning)
AA-Omniscience Accuracy	63.5%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Omniscience Accuracy	64.5%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Omniscience Accuracy	64.6%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Omniscience Accuracy	65.4%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Omniscience Accuracy	66.2%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—
AA-Omniscience Hallucination Rate	67.6%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page)
AA-Omniscience Hallucination Rate	68.4%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page)
AA-Omniscience Hallucination Rate	67.6%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page)
AA-Omniscience Hallucination Rate	65.7%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page)
AA-Omniscience Hallucination Rate	58.6%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page)
AA-Omniscience Index	38.9	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised
AA-Omniscience Index	40.3	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised
AA-Omniscience Index	40.6	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised
AA-Omniscience Index	42.7	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised
AA-Omniscience Index	46.4	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised
ARC-AGI-2	87.5%	MediumClaude Opus 5.5 (Medium)	Independent testIndependentARC Prize ↗	30 Sep 2026	—	semi-private set; $0.34/task; data https://arcprize.org/media/data/leaderboard/v2.json
ARC-AGI-2	93.3%	HighClaude Opus 5.5 (High)	Independent testIndependentARC Prize ↗	30 Sep 2026	—	semi-private set; $0.41/task; data https://arcprize.org/media/data/leaderboard/v2.json
ARC-AGI-2	92.5%	Extra highClaude Opus 5.5 (XHigh)	Independent testIndependentARC Prize ↗	30 Sep 2026	—	semi-private set; $0.67/task; data https://arcprize.org/media/data/leaderboard/v2.json
ARC-AGI-2	91.7%	MaxClaude Opus 5.5 (Max)	Independent testIndependentARC Prize ↗	30 Sep 2026	—	semi-private set; $1.85/task; data https://arcprize.org/media/data/leaderboard/v2.json
Artificial Analysis Coding Agent Index	66	MaxOpus 5.5 (max)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	Claude Code	agent Claude Code; components: DeepSWE v1.1 68.4, SWE-Atlas-QnA 66.4, Terminal-Bench v4 63.1; avg cost $13.04/task; avg wall time 64 min/task
Artificial Analysis Intelligence Index	42.3	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA slug claude-opus-5-5-low; list price $4/20 per 1M in/out; cost to run AA Intelligence Index $0.55/task
Artificial Analysis Intelligence Index	51.2	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA slug claude-opus-5-5-medium; list price $4/20 per 1M in/out; cost to run AA Intelligence Index $1.34/task
Artificial Analysis Intelligence Index	53.6	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA slug claude-opus-5-5-high; list price $4/20 per 1M in/out; cost to run AA Intelligence Index $1.82/task
Artificial Analysis Intelligence Index	56	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA slug claude-opus-5-5-xhigh; list price $4/20 per 1M in/out; cost to run AA Intelligence Index $3.46/task
Artificial Analysis Intelligence Index	57.6	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA slug claude-opus-5-5; list price $4/20 per 1M in/out; cost to run AA Intelligence Index $5.98/task
Artificial Analysis output speed	72 tok/s	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	tokens/sec (median output speed, first-party API); TTFT 13.5s; list price $4/20 per 1M in/out
Artificial Analysis output speed	71 tok/s	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	tokens/sec (median output speed, first-party API); TTFT 24.4s; list price $4/20 per 1M in/out
Artificial Analysis output speed	73 tok/s	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	tokens/sec (median output speed, first-party API); TTFT 38.9s; list price $4/20 per 1M in/out
Artificial Analysis output speed	77 tok/s	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	tokens/sec (median output speed, first-party API); TTFT 137.2s; list price $4/20 per 1M in/out
Artificial Analysis output speed	90 tok/s	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	tokens/sec (median output speed, first-party API); TTFT 724.5s; list price $4/20 per 1M in/out
ArXivMath (no tools)	91.2%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	ArXivMath Aug 2026 (57 problems), no tools, avg of 4
ArXivMath (with tools)	96.9%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	Aug 2026, code execution
AutomationBench	24.2%	Loweffort=low	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	AutomationBench 1.0.6 (Zapier); vendor-estimated API cost/task $0.51; competitor number as plotted by OpenAI ("Opus 5.5 w/ fallbacks")
AutomationBench	23.3%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	AutomationBench (Zapier; end-to-end business workflows across connected apps); run by Zapier without fallback models (safeguard interventions counted as failures); vendor-reported cost/task $0.5
AutomationBench	29.5%	Mediumeffort=medium	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	AutomationBench 1.0.6 (Zapier); vendor-estimated API cost/task $0.65; competitor number as plotted by OpenAI ("Opus 5.5 w/ fallbacks")
AutomationBench	28.6%	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	AutomationBench (Zapier; end-to-end business workflows across connected apps); run by Zapier without fallback models (safeguard interventions counted as failures); vendor-reported cost/task $0.64
AutomationBench	33.0%	Higheffort=high	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	AutomationBench 1.0.6 (Zapier); vendor-estimated API cost/task $0.71; competitor number as plotted by OpenAI ("Opus 5.5 w/ fallbacks")
AutomationBench	32.0%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	AutomationBench (Zapier; end-to-end business workflows across connected apps); run by Zapier without fallback models (safeguard interventions counted as failures); vendor-reported cost/task $0.7
AutomationBench	35.8%	Extra higheffort=xhigh	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	AutomationBench 1.0.6 (Zapier); vendor-estimated API cost/task $0.89; competitor number as plotted by OpenAI ("Opus 5.5 w/ fallbacks")
AutomationBench	34.4%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	AutomationBench (Zapier; end-to-end business workflows across connected apps); run by Zapier without fallback models (safeguard interventions counted as failures); vendor-reported cost/task $0.86
AutomationBench	40.0%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	AutomationBench (Zapier; end-to-end business workflows across connected apps); run by Zapier without fallback models (safeguard interventions counted as failures); vendor-reported cost/task $1.37
Chartography (no tools)	64.4%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	28 Sep 2026	—	no tools
Chartography (with tools)	89.0%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	Chartography (Surge AI chart-reading) with tools
CursorBench	43.7%	LowOpus 5.5 Low	Independent testIndependentCursor (CursorBench) ↗	1 Oct 2026	Cursor agent	CursorBench 4.0 (tasks from real Cursor sessions; v4.0 introduced 2026-09-10); rank 18; $1.17/task; 15,811 tokens/task; 28 steps/task
CursorBench	43.7%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Cursor agent	CursorBench 4.0, run end-to-end by Cursor in its production agent harness; Opus 5.5 cost estimated by Anthropic from Cursor token counts; vendor-reported cost/task $1.18
CursorBench	52.5%	MediumOpus 5.5 Medium	Independent testIndependentCursor (CursorBench) ↗	1 Oct 2026	Cursor agent	CursorBench 4.0 (tasks from real Cursor sessions; v4.0 introduced 2026-09-10); rank 6; $2.91/task; 37,954 tokens/task; 54 steps/task
CursorBench	52.5%	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Cursor agent	CursorBench 4.0, run end-to-end by Cursor in its production agent harness; Opus 5.5 cost estimated by Anthropic from Cursor token counts; vendor-reported cost/task $2.9
CursorBench	56.0%	HighOpus 5.5 High	Independent testIndependentCursor (CursorBench) ↗	1 Oct 2026	Cursor agent	CursorBench 4.0 (tasks from real Cursor sessions; v4.0 introduced 2026-09-10); rank 3; $3.97/task; 53,078 tokens/task; 68 steps/task
CursorBench	56.0%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Cursor agent	CursorBench 4.0, run end-to-end by Cursor in its production agent harness; Opus 5.5 cost estimated by Anthropic from Cursor token counts; vendor-reported cost/task $3.97
CursorBench	56.0%	Extra highOpus 5.5 Extra High	Independent testIndependentCursor (CursorBench) ↗	1 Oct 2026	Cursor agent	CursorBench 4.0 (tasks from real Cursor sessions; v4.0 introduced 2026-09-10); rank 2; $6.98/task; 101,083 tokens/task; 109 steps/task
CursorBench	56.0%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Cursor agent	CursorBench 4.0, run end-to-end by Cursor in its production agent harness; Opus 5.5 cost estimated by Anthropic from Cursor token counts; vendor-reported cost/task $6.99
CursorBench	57.8%	MaxOpus 5.5 Max	Independent testIndependentCursor (CursorBench) ↗	1 Oct 2026	Cursor agent	CursorBench 4.0 (tasks from real Cursor sessions; v4.0 introduced 2026-09-10); rank 1; $13.43/task; 218,363 tokens/task; 185 steps/task
CursorBench	57.8%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Cursor agent	CursorBench 4.0, run end-to-end by Cursor in its production agent harness; Opus 5.5 cost estimated by Anthropic from Cursor token counts; vendor-reported cost/task $13.43
DeepSWE	68.4%	MaxOpus 5.5 (max)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	Claude Code	AA Coding Agent Index component; pass@1 avg of 3 attempts
DeepSWE	74.2%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
Design Arena (all categories)	1399	Defaultclaude-opus-5-5	Independent testIndependentDesign Arena ↗	1 Oct 2026	—	Bradley-Terry Elo; ±10.4 SE; 1259 battles; win rate 65.3%
Epoch Capabilities Index	167.3	Default	Independent testIndependentEpoch AI ↗	1 Oct 2026	—	Epoch Capabilities Index; 90% CI 164.0-172.0; best of listed model versions
EQ-Bench Creative Writing v3 (Elo)	2050	Default*claude-opus-5-5	Independent testIndependentEQ-Bench ↗	1 Oct 2026	—	leaderboard rank 7; rubric score 16.79/20; slop 10.43; avg length 6041 chars; Elo judged by Claude Sonnet 4.6; effort not stated by EQ-Bench; marked new (*)
FrontierCode	47.3%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Main (150 tasks, run & scored by Cognition; composite of correctness + mergeability rubric; penalizes out-of-scope edits); Anthropic: scores decline above medium and mostly recover at max (out-of-scope changes penalized); vendor-reported cost/task $0.4
FrontierCode	54.6%	Mediumclaude-opus-5-5_medium	Independent testIndependentFrontierCode (Cognition) via Epoch AI ↗	1 Oct 2026	Claude Code	read from Epoch AI benchmark_data.zip (frontiercode_external.csv); original leaderboard https://cognition.com/frontiercode; harness claude-code; Mean@5
FrontierCode	54.6%	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Main (150 tasks, run & scored by Cognition; composite of correctness + mergeability rubric; penalizes out-of-scope edits); Anthropic: scores decline above medium and mostly recover at max (out-of-scope changes penalized); vendor-reported cost/task $0.8
FrontierCode	54.0%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Main (150 tasks, run & scored by Cognition; composite of correctness + mergeability rubric; penalizes out-of-scope edits); Anthropic: scores decline above medium and mostly recover at max (out-of-scope changes penalized); vendor-reported cost/task $1.09
FrontierCode	51.4%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Main (150 tasks, run & scored by Cognition; composite of correctness + mergeability rubric; penalizes out-of-scope edits); Anthropic: scores decline above medium and mostly recover at max (out-of-scope changes penalized); vendor-reported cost/task $2.25
FrontierCode	54.4%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Main (150 tasks, run & scored by Cognition; composite of correctness + mergeability rubric; penalizes out-of-scope edits); Anthropic: scores decline above medium and mostly recover at max (out-of-scope changes penalized); vendor-reported cost/task $6.19
FrontierCode v1.1 (Extended)	65.3%	Mediumadaptive thinking, effort=medium (best)	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code	FrontierCode v1.1 Extended; declines above medium, mostly recovers at max
FrontierCode v1.1 (Extended)	63.6%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code
FrontierMath (Tiers 1-3)	91.2%	Maxclaude-opus-5-5_max	Independent testIndependentEpoch AI ↗	22 Sep 2026	—	FrontierMath Tiers 1-3 (v2), Epoch-run; stderr 1.7pt; data https://epoch.ai/data/benchmark_data.zip (frontiermath_tiers_1_3_v2.csv)
FrontierMath Tier 4	95.0%	Maxclaude-opus-5-5_max	Independent testIndependentEpoch AI ↗	22 Sep 2026	—	FrontierMath Tier 4 (v2), Epoch-run; stderr 4.1pt; data https://epoch.ai/data/benchmark_data.zip (frontiermath_tier_4_v2.csv)
FrontierSWE	62.3%	Default	Independent testIndependentFrontierSWE ↗	1 Oct 2026	proximus	FrontierSWE V2, mean@5 over 34 tasks (20h budget); ±9.8; $98.87/trial; 14.2h/trial; 'Per provider' view shows best entry per provider; Epoch notes runs use max reasoning effort
FrontierSWE v2	62.3%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Proximal harness	avg of 5 trials unless otherwise noted
GDP.pdf	25.6%	Loweffort=low	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	GDP.pdf (Surge AI; professional questions over complex PDFs); vendor-estimated API cost/task $0.76412; competitor number as plotted by OpenAI ("Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback")
GDP.pdf	25.6%	Mediumeffort=medium	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	GDP.pdf (Surge AI; professional questions over complex PDFs); vendor-estimated API cost/task $0.79532; competitor number as plotted by OpenAI ("Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback")
GDP.pdf	28.8%	Higheffort=high	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	GDP.pdf (Surge AI; professional questions over complex PDFs); vendor-estimated API cost/task $0.82543; competitor number as plotted by OpenAI ("Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback")
GDP.pdf	26.6%	Extra higheffort=xhigh	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	GDP.pdf (Surge AI; professional questions over complex PDFs); vendor-estimated API cost/task $0.96427; competitor number as plotted by OpenAI ("Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback")
GDP.pdf	26.2%	Maxeffort=max	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	GDP.pdf (Surge AI; professional questions over complex PDFs); vendor-estimated API cost/task $1.55234; competitor number as plotted by OpenAI ("Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback")
GDPval-AA v2.1	1224	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	GDPval-AA v2.1 (Artificial Analysis, Elo from blind pairwise comparisons, 44 occupations); vendor-reported cost/task $0.21
GDPval-AA v2.1	1576	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	GDPval-AA v2.1 (Artificial Analysis, Elo from blind pairwise comparisons, 44 occupations); vendor-reported cost/task $0.86
GDPval-AA v2.1	1692	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	GDPval-AA v2.1 (Artificial Analysis, Elo from blind pairwise comparisons, 44 occupations); vendor-reported cost/task $1.54
GDPval-AA v2.1	1820	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	95% CI 1798.02-1842.19
GDPval-AA v2.1	1820	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	GDPval-AA v2.1 (Artificial Analysis, Elo from blind pairwise comparisons, 44 occupations); vendor-reported cost/task $4.21
GDPval-AA v2.1	1846	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	95% CI 1822.93-1869.4
GDPval-AA v2.1	1846	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	GDPval-AA v2.1 (Artificial Analysis, Elo from blind pairwise comparisons, 44 occupations); vendor-reported cost/task $8.92
Harvey LAB-AA	91.2%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	criteria pass rate, AA-run
HealthBench Professional	65.6%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
HLE Diamond	55.0%	Defaultclaude-opus-5-5	Independent testIndependentScale AI SEAL ↗	1 Oct 2026	—	rank 2 (Scale rank accounts for CI); ±3.1; entry added 2026-09-03
Humanity's Last Exam	48.3%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-low)
Humanity's Last Exam	54.7%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-medium)
Humanity's Last Exam	55.6%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-high)
Humanity's Last Exam	57.5%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-xhigh)
Humanity's Last Exam	61.4%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5)
Humanity's Last Exam	64.4%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
Humanity's Last Exam (with tools)	67.7%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	with web search, web fetch, programmatic tool calling, code execution; 1M total token cap
IOI (Vals)	95.1%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	—	±4.944 stderr; $5.258/test
LLM Creative Story-Writing Benchmark (Lech Mazur)	3.8	HighClaude Opus 5.5 (high)	Independent testIndependentLech Mazur (LLM Creative Story-Writing Benchmark) ↗	28 Sep 2026	—	rank 3/56; Thurstone comparison score (centered at 0); est. win chance 93%; 95% bootstrap 3.676 to 3.882
LMArena Code Arena (WebDev)	1818	Maxclaude-opus-5.5-max	Independent testIndependentLMArena ↗	30 Sep 2026	—	Code Arena \| WebDev overall (agentic web-dev, raw); rank 1 (CI rank 1-1); 95% CI 1801-1834; 1976 votes
LMArena Text - Coding category	1538	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	text arena coding category, style control on; rank 11 (CI rank 1-50); 95% CI 1519-1557; 918 votes
LMArena Text - Creative Writing	1515	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 2 (rank range 1-7); 95% CI 1493.9-1536.1; 908 votes; style-controlled
LMArena Text - Expert	1539	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 4 (rank range 1-43); 95% CI 1511.1-1566.8; 448 votes; style-controlled
LMArena Text - Hard Prompts	1533	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 3 (rank range 1-11); 95% CI 1521.1-1545.3; 2437 votes; style-controlled
LMArena Text - Instruction Following	1514	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 2 (rank range 1-10); 95% CI 1498.2-1530.7; 1337 votes; style-controlled
LMArena Text - Longer Query	1528	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 2 (rank range 1-9); 95% CI 1512.9-1542.7; 1708 votes; style-controlled
LMArena Text - Multi-Turn	1497	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 11 (rank range 2-75); 95% CI 1470.6-1523.1; 496 votes; style-controlled
LMArena Text - Non-English	1496	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 2 (rank range 1-16); 95% CI 1484.2-1508.1; 2470 votes; style-controlled
LMArena Text - Occupational: Business, Management & Finance	1493	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 10 (rank range 1-55); 95% CI 1470.7-1515.0; 716 votes; style-controlled
LMArena Text - Occupational: Legal & Government	1490	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 23 (rank range 1-100); 95% CI 1457.7-1523.0; 336 votes; style-controlled
LMArena Text - Occupational: Writing, Literature & Language	1519	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	rank 2 (rank range 1-7); 95% CI 1500.8-1537.3; 1159 votes; style-controlled
LMArena Text (overall)	1504	Highclaude-opus-5.5-high	Independent testIndependentLMArena ↗	30 Sep 2026	—	text overall, style control on; rank 4 (CI rank 2-14); 95% CI 1494-1514; 3932 votes
OfficeQA Pro	67.7%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
OSWorld 2.1 (partial score)	81.8%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	partial score; system card table labels it OSWorld 2.0 (partial/strict 81.8/48.7)
OSWorld 2.1 (strict pass rate)	48.7%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	strict pass rate (card labels OSWorld 2.0)
PRBench Legal (Scale)	47.6%	Defaultclaude-opus-5-5	Independent testIndependentScale AI (SEAL) ↗	25 Sep 2026	—	Scale rank 8; ±2.2 CI
ProgramBench (fully resolved)	18.5%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	mini-SWE-agent	±2.753 stderr; $68.051/test; strict fully-resolved rate
ProgramBench (fully resolved)	91.2%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	mini-swe-agent	avg of 5 trials unless otherwise noted
SciCode	58.6%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-low)
SciCode	59.3%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-medium)
SciCode	60.4%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-high)
SciCode	65.0%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-xhigh)
SciCode	66.9%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5)
SimpleBench	88.4%	DefaultClaude Opus 5.5	Independent testIndependentSimpleBench ↗	24 Sep 2026	—	AVG@5, temp 0.7; rank 1st; human baseline 83.7%; data https://simple-bench.com/static/js/leaderboard-data.js
SimpleQA Verified	72.2%	Maxclaude-opus-5-5_max	Independent testIndependentEpoch AI ↗	22 Sep 2026	—	Epoch-run (no tools); ±1.42 stderr
SWE Atlas - Codebase QnA	66.4%	MaxOpus 5.5 (max)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	Claude Code	AA Coding Agent Index component; pass@1 avg of 3 attempts
SWE-bench Multilingual	93.9%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
SWE-bench Multimodal	61.4%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
SWE-bench Pro (Anthropic internal subset)	87.4%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	1 Oct 2026	—	Anthropic internal 478-problem SWE-bench Pro subset (not comparable to public leaderboard), priced as billed; $0.12 per solved task
SWE-bench Pro (Anthropic internal subset)	92.8%	Mediumadaptive thinking, effort=medium (default)	Maker's own figureVendor-reportedAnthropic ↗	1 Oct 2026	—	Anthropic internal 478-problem SWE-bench Pro subset (not comparable to public leaderboard), priced as billed; $0.22 per solved task; ~2.5 pts below high for ~70% of the cost
SWE-bench Pro (Anthropic internal subset)	95.3%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	1 Oct 2026	—	Anthropic internal 478-problem SWE-bench Pro subset (not comparable to public leaderboard), priced as billed; $0.29 per task; xhigh adds ~1.4 pts for 2.5x the cost of high
SWE-Bench Pro (public, v1)	89.9%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
Terminal-Bench 2.1	87.6%	Higheffort=high	Independent testIndependentVals.ai ↗	27 Sep 2026	Terminus 2	±1.716 stderr; $0.496/test
Terminal-Bench 4.0	31.3%	LowClaude Opus 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-low)
Terminal-Bench 4.0	38.5%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench 4.0 (66 tasks); Anthropic internal runs, 5-15 trials/task; xhigh is the reported headline (max within noise, SE ±2.6); production safeguards on, fallbacks on 2.5% of requests; vendor-reported cost/task $1.29
Terminal-Bench 4.0	52.5%	MediumClaude Opus 5.5 (Adaptive Reasoning, Medium Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-medium)
Terminal-Bench 4.0	57.6%	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench 4.0 (66 tasks); Anthropic internal runs, 5-15 trials/task; xhigh is the reported headline (max within noise, SE ±2.6); production safeguards on, fallbacks on 2.5% of requests; vendor-reported cost/task $2.94
Terminal-Bench 4.0	56.6%	HighClaude Opus 5.5 (Adaptive Reasoning, High Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-high)
Terminal-Bench 4.0	64.2%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench 4.0 (66 tasks); Anthropic internal runs, 5-15 trials/task; xhigh is the reported headline (max within noise, SE ±2.6); production safeguards on, fallbacks on 2.5% of requests; vendor-reported cost/task $3.88
Terminal-Bench 4.0	59.6%	Extra highClaude Opus 5.5 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5-xhigh)
Terminal-Bench 4.0	66.4%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench 4.0 (66 tasks); Anthropic internal runs, 5-15 trials/task; xhigh is the reported headline (max within noise, SE ±2.6); production safeguards on, fallbacks on 2.5% of requests; vendor-reported cost/task $7.35
Terminal-Bench 4.0	65.2%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	mini-SWE-agent	±0 stderr; $13.199/test
Terminal-Bench 4.0	63.1%	MaxOpus 5.5 (max)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	Claude Code	AA Coding Agent Index component run in the vendor agent harness; pass@1 avg of 3 attempts
Terminal-Bench 4.0	59.6%	MaxClaude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)	Independent testIndependentArtificial Analysis ↗	1 Oct 2026	—	AA-run evaluation (AA slug claude-opus-5-5)
Terminal-Bench 4.0	64.8%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench 4.0 (66 tasks); Anthropic internal runs, 5-15 trials/task; xhigh is the reported headline (max within noise, SE ±2.6); production safeguards on, fallbacks on 2.5% of requests; vendor-reported cost/task $11.24
Terminal-Bench-Science 0.1	63.3%	Maxeffort=max	Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗	29 Sep 2026	—	Terminal-Bench Science 0.1; vendor-estimated API cost/task $23.2108; competitor number as plotted by OpenAI
Terminal-Bench-Science 0.1	58.7%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	Claude Code (--bare)	Terminal-Bench-Science 0.1 (70 tasks); SE ±4.8; safeguards on (fallbacks on 3.9% of requests)
Toolathlon	77.8%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	avg of 5 trials unless otherwise noted
Vals Code Migration	66.7%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	—	±4.329 stderr; $112.971/test
Vals Index	67.0%	Maxeffort=max	Independent testIndependentVals.ai ↗	30 Sep 2026	—	±0.888 stderr; $32.144/test
Vals Legal Research Bench	50.5%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	—	vals id anthropic/claude-opus-5-5; rank 5/72; ±3.475 stderr; $25.163725/test
Vals Public Benefits Bench v1.1	70.6%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	—	vals id anthropic/claude-opus-5-5; rank 3/45; ±1.185 stderr; $5.952407/test
Vals SRE Bench	33.6%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	—	±2.923 stderr; $34.654/test
Vals Vibe Code Bench	90.3%	Maxeffort=max	Independent testIndependentVals.ai ↗	29 Sep 2026	OpenHands	±1.526 stderr; $57.923/test
Vending-Bench 2	$9,235	Default	Independent testIndependentAndon Labs ↗	1 Oct 2026	—	final money balance after simulated year, arithmetic mean across runs; ±$785; rank 8; only top 10 rendered server-side (57 more behind 'Show more')
WANDR	31.2%	Lowadaptive thinking, effort=low	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	WANDR (Perplexity wide-search benchmark), soft F1; Anthropic-modified setup (offline search/fetch, 980k token budget), not comparable to Perplexity numbers; vendor-reported cost/task $1.2
WANDR	62.8%	Mediumadaptive thinking, effort=med	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	WANDR (Perplexity wide-search benchmark), soft F1; Anthropic-modified setup (offline search/fetch, 980k token budget), not comparable to Perplexity numbers; vendor-reported cost/task $11.2
WANDR	67.3%	Highadaptive thinking, effort=high	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	WANDR (Perplexity wide-search benchmark), soft F1; Anthropic-modified setup (offline search/fetch, 980k token budget), not comparable to Perplexity numbers; vendor-reported cost/task $16.88
WANDR	71.3%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	WANDR (Perplexity wide-search benchmark), soft F1; Anthropic-modified setup (offline search/fetch, 980k token budget), not comparable to Perplexity numbers; vendor-reported cost/task $29.06
WANDR	72.3%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	22 Sep 2026	—	WANDR (Perplexity wide-search benchmark), soft F1; Anthropic-modified setup (offline search/fetch, 980k token budget), not comparable to Perplexity numbers; vendor-reported cost/task $37.92