Language modelActive

Claude Opus 4.6

Anthropic

Released: February 2026
Data date: September 8, 2026
Predecessor: Claude Opus 4.5
Successor: Claude Opus 4.7

Speed: 39.2 tokens/s
Output throughput
Price: $5 · $25
Input · Output / 1M tokens
Input: Text, Image
Output: Text

Claude Opus 4.6 was the first Opus model with a one-million-token context window. It followed Claude Opus 4.5 and preceded Claude Opus 4.7. Anthropic also launched it alongside API context compaction and agent teams in Claude Code.

For coding, Anthropic highlighted work across large codebases, code review, and debugging. Its 1M-token context window, then in beta, could hold far more material in a request for long tasks.

Specifications and access

Specification	Value and source
Weights	ProprietarySource
Access	APISource
Context window	1,000,000 tokensSource
Knowledge cutoff	May 2025Source
Notes	1M context, 128K output, agent teams, SWE-bench ~80.8%Source
API model ID	claude-opus-4-6Source
Source status	active (legacy)Source
Release	2026-02-05Source
Maximum output	128,000 tokensSource
Input	Text, imagesSource
Output	TextSource
Reasoning mode	adaptive; extended legacy modeSource
Knowledge cutoff	2025-05 reliable / 2025-08 trainingSource
Tool calling	YesSource
Structured output	YesSource

Published benchmarks

Humanity's Last Exam

Overall

2,500 difficult, partly multimodal expert questions across more than 100 subjects. The benchmark measures closed-ended expert knowledge and reasoning, not open research or general intelligence.

1.Gemini 3.1 Pro Preview46.44%

2.GPT-5.4 Pro44.32%

3.GPT-5.436.24%

4.Claude Opus 4.736.2%

5.Claude Opus 4.634.44%

6.GPT-5 Pro31.64%

7.GPT-525.32%

higher is betterAs of September 2, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: hle-final-scale-public-temperature-0
Metric: accuracy
Scale: 0-100
Participants: 7
Methodology and context: All 2,500 public questions, temperature 0, and an automatic answer extractor.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Model snapshot: claude-opus-4-6-thinking-max. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 34.44% ± 1.86. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-09-02-9e661b4b9e47. Source payload hash: 9e661b4b9e47fc78acc41dd3d83ae7363672deeb05d38a32fd0dcc88c7502935. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

Agent tasks

METR Time Horizon

Human task duration at which a model-agent system reaches a 50% or 80% predicted success probability. Minutes refer to human task duration, not model runtime. METR considers estimates above 16 hours unreliable.

Sub-task	Claude Opus 4.6 (This model)	GPT-5.4	Claude Opus 4.5
50% time horizon	718.81 (best value in this row)	341.74	292.99
80% time horizon	69.87 (best value in this row)	53.88	49.43

Unit: minuteshigher is betterAs of August 26, 2026Source: METRbars scaled within each row, not across rows

MultiChallenge

Overall

Multi-turn conversations test memory, instruction retention, versioned editing, and self-coherence. Tasks were selected from shared failures of six older frontier models and may disadvantage those systems.

1.Gemini 3.1 Pro Preview71.37%

2.GPT-5.4 Pro69.23%

3.GPT-563.19%

4.o3-pro62.4%

5.Claude Opus 4.558.97%

6.o356.62%

7.Claude Opus 4.656.02%

8.Claude Sonnet 4.555.32%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-multichallenge-overall
Metric: rubric pass rate
Scale: 0-100
Participants: 8
Methodology and context: Multi-turn conversations with binary task-specific rubrics.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 56.02% ± 1.08. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-c7bc8d4b82b8. Source payload hash: c7bc8d4b82b8a93b33209cb8b5c8da53dc01bac7aa0cd5ca76c4ae365619525d. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

PRBench Finance

Overall

600 realistic finance questions with weighted criteria, including a 300-question hard subset. An o4-mini judge scores the responses, so domain correctness and operational reliability remain separate questions.

1.Claude Fable 553.86%

2.Claude Opus 4.653.28%

3.GPT-551.32%

4.GPT-5 Pro51.06%

5.GPT-5.6 Sol50.45%

6.o3-pro49.08%

7.o347.69%

8.Claude Opus 4.546.16%

9.GPT-5.445.63%

10.Claude Sonnet 4.543.79%

11.Gemini 3.1 Pro Preview41.87%

12.o4-mini39.22%

13.Gemini 2.5 Pro38.92%

14.Gemini 2.5 Flash38.41%

15.GPT-4.134.32%

16.Llama 4 Maverick22.36%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-prbench-finance-full
Metric: weighted rubric score
Scale: 0-100
Participants: 16
Methodology and context: 600 finance tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 53.28% ± 0.18. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-1196d0ba7a4c. Source payload hash: 1196d0ba7a4c8d626776c467c1d9d53e493e848fbce8f12f327c3864f83ac2c4. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

PRBench Legal

Overall

500 realistic legal questions with weighted criteria across jurisdictions and practice areas. The benchmark measures professional response quality but cannot replace legal review or human accountability.

1.Claude Fable 552.56%

2.Claude Opus 4.652.27%

3.GPT-5.6 Sol50.5%

4.GPT-5 Pro49.89%

5.o3-pro49.67%

6.GPT-548.96%

7.o348.57%

8.GPT-5.444.35%

9.Claude Opus 4.544.21%

10.Gemini 3.1 Pro Preview44.02%

11.Gemini 2.5 Pro41.43%

12.Gemini 2.5 Flash41.02%

13.Claude Sonnet 4.540.84%

14.o4-mini38.11%

15.GPT-4.136.48%

16.Llama 4 Maverick24.84%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-prbench-legal-full
Metric: weighted rubric score
Scale: 0-100
Participants: 16
Methodology and context: 500 legal tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 52.27% ± 0.66. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-39dd22178287. Source payload hash: 39dd221782879e08cf3582225a60f6e91432e1e8d4968e2fb60b53c917fafe6a. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

VisualToolBench

Overall

1,204 visual tasks require image manipulation, Python, web search, or calculator use with at most 20 tool calls. The score belongs to the complete multimodal tool system and cannot isolate the model alone.

1.GPT-5.429.17%

2.Gemini 3.1 Pro Preview28.97%

3.Claude Opus 4.627.52%

4.o313.74%

5.Gemini 2.5 Pro11.75%

6.o4-mini11.12%

7.GPT-4.15.52%

8.Gemini 2.5 Flash4.69%

9.Llama 4 Scout1.58%

10.Llama 4 Maverick1.41%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-visual-toolbench-apr
Metric: average pass rate
Scale: 0-100
Participants: 10
Methodology and context: 1,204 tasks, six tools, and at most 20 tool calls per task.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: thinking. Score display: 27.52% ± 1.45. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-eb72202a59a1. Source payload hash: eb72202a59a1301462dd728b19363526d0f1ed2c2b58120de39b0027aafc74ac. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

Cybersecurity

SecCodeBench-V2

98 executable project cases across 22 CWE categories test functionality before security. The weighted score combines generation, repair, native prompts, and security-aware prompts.

Sub-task	Claude Opus 4.6 (This model)	Claude Opus 4.5	Qwen 3.7 Max	GPT-5.4
Repair without hints	73.1%	73.67% (best value in this row)	72.5%	61.41%
Repair with security hints	81.73%	82.18%	86.77% (best value in this row)	80.04%
Generation without hints	51.05%	58.02% (best value in this row)	55.12%	50.97%
Generation with security hints	70.65%	72.1% (best value in this row)	70.85%	67.82%
Weighted total	64.9%	68.1% (best value in this row)	66.81%	59.74%

higher is betterAs of August 9, 2026Source: SecCodeBenchbars scaled within each row, not across rows

Recorded measurement details

Humanity's Last Exam: Overall (Show measurement, test conditions, and source)

Source value: 34.44
Score: 34.44
Metric: accuracy
Unit: %
Benchmark version: final 2,500-question public set
Category: knowledge
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: hle-final-scale-public-temperature-0
Confidence interval: 1.86
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-09-02
Methodology: All 2,500 public questions, temperature 0, and an automatic answer extractor.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Model snapshot: claude-opus-4-6-thinking-max. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 34.44% ± 1.86. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-09-02-9e661b4b9e47. Source payload hash: 9e661b4b9e47fc78acc41dd3d83ae7363672deeb05d38a32fd0dcc88c7502935. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: All 2,500 public questions, temperature 0, and an automatic answer extractor.

Source Scale AI

METR Time Horizon: 50% time horizon (Show measurement, test conditions, and source)

Source value: 718.80683
Score: 718.80683
Metric: 50% time horizon
Unit: minutes
Benchmark version: TH 1.1
Category: agentic
Task: p50
Task label: 50% time horizon
Direction: Higher is better
Comparison cohort: metr-th1-1-p50
Source type: independent-evaluator
Evaluator: METR
Status: active
Retrieved at: 2026-08-26
Methodology: ReAct agent with six independent runs per task. Time is estimated human completion duration.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Model snapshot: claude-opus-4-6-2026-02-05. Evaluator: METR. Dataset type: public. Leaderboard date: 2026-05-08. Harness: ReAct. Agent: ReAct. Attempts: 6 independent runs per task. Score display: 12 h (95% confidence interval 5.3 h to 60.6 h). Source ID: metr. Evidence capture ID: curated-metr-2026-08-26-1ca8b69f5a82. Source payload hash: 1ca8b69f5a826f53e175e4b094b9f4b51e14a39ba2c084eab283897836d49d20. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: ReAct agent with six independent runs per task. Time is estimated human completion duration.

Source METR

METR Time Horizon: 80% time horizon (Show measurement, test conditions, and source)

Source value: 69.874587
Score: 69.874587
Metric: 80% time horizon
Unit: minutes
Benchmark version: TH 1.1
Category: agentic
Task: p80
Task label: 80% time horizon
Direction: Higher is better
Comparison cohort: metr-th1-1-p80
Source type: independent-evaluator
Evaluator: METR
Status: active
Retrieved at: 2026-08-26
Methodology: ReAct agent with six independent runs per task. Time is estimated human completion duration.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Model snapshot: claude-opus-4-6-2026-02-05. Evaluator: METR. Dataset type: public. Leaderboard date: 2026-05-08. Harness: ReAct. Agent: ReAct. Attempts: 6 independent runs per task. Score display: 1.2 h (95% confidence interval 27 min to 2.8 h). Source ID: metr. Evidence capture ID: curated-metr-2026-08-26-318bdb5530ca. Source payload hash: 318bdb5530cad90be3765939e5139aa3ed2f5a6b9e463cf3398f516e4710e864. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: ReAct agent with six independent runs per task. Time is estimated human completion duration.

Source METR

MultiChallenge: Overall (Show measurement, test conditions, and source)

Source value: 56.02
Score: 56.02
Metric: rubric pass rate
Unit: %
Benchmark version: four multi-turn challenge types
Category: productivity
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-multichallenge-overall
Confidence interval: 1.08
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: Multi-turn conversations with binary task-specific rubrics.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 56.02% ± 1.08. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-c7bc8d4b82b8. Source payload hash: c7bc8d4b82b8a93b33209cb8b5c8da53dc01bac7aa0cd5ca76c4ae365619525d. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: Multi-turn conversations with binary task-specific rubrics.

Source Scale AI

PRBench Finance: Overall (Show measurement, test conditions, and source)

Source value: 53.28
Score: 53.28
Metric: weighted rubric score
Unit: %
Benchmark version: full finance set
Category: finance
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-prbench-finance-full
Confidence interval: 0.18
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 600 finance tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 53.28% ± 0.18. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-1196d0ba7a4c. Source payload hash: 1196d0ba7a4c8d626776c467c1d9d53e493e848fbce8f12f327c3864f83ac2c4. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 600 finance tasks with task-specific rubrics and o4-mini as judge.

Source Scale AI

PRBench Legal: Overall (Show measurement, test conditions, and source)

Source value: 52.27
Score: 52.27
Metric: weighted rubric score
Unit: %
Benchmark version: full legal set
Category: legal
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-prbench-legal-full
Confidence interval: 0.66
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 500 legal tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: none. Score display: 52.27% ± 0.66. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-39dd22178287. Source payload hash: 39dd221782879e08cf3582225a60f6e91432e1e8d4968e2fb60b53c917fafe6a. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 500 legal tasks with task-specific rubrics and o4-mini as judge.

Source Scale AI

VisualToolBench: Overall (Show measurement, test conditions, and source)

Source value: 27.52
Score: 27.52
Metric: average pass rate
Unit: %
Benchmark version: 1,204-task release
Category: multimodal
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-visual-toolbench-apr
Confidence interval: 1.45
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 1,204 tasks, six tools, and at most 20 tool calls per task.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Reasoning effort: thinking. Score display: 27.52% ± 1.45. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-eb72202a59a1. Source payload hash: eb72202a59a1301462dd728b19363526d0f1ed2c2b58120de39b0027aafc74ac. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 1,204 tasks, six tools, and at most 20 tool calls per task.

Source Scale AI

SecCodeBench-V2: Repair without hints (Show measurement, test conditions, and source)

Source value: 73.1
Score: 73.1
Metric: native repair pass@1
Unit: %
Benchmark version: 2.2.0
Category: cybersecurity
Task: fix
Task label: Repair without hints
Direction: Higher is better
Comparison cohort: seccodebench-v2-2-0:official:fix
Source type: official-leaderboard
Evaluator: SecCodeBench
Status: active
Retrieved at: 2026-08-09
Methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: SecCodeBench. Dataset type: public. Leaderboard date: 2026-02-16. Source ID: seccodebench. Evidence capture ID: curated-seccodebench-2026-08-09-cbeb9cb3730c. Source payload hash: cbeb9cb3730ca29af3d8b7a5e2fca139765542955a6bbf61409536cafac54c77. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.

Source SecCodeBench

SecCodeBench-V2: Repair with security hints (Show measurement, test conditions, and source)

Source value: 81.73
Score: 81.73
Metric: security-aware repair pass@1
Unit: %
Benchmark version: 2.2.0
Category: cybersecurity
Task: fixHints
Task label: Repair with security hints
Direction: Higher is better
Comparison cohort: seccodebench-v2-2-0:official:fixHints
Source type: official-leaderboard
Evaluator: SecCodeBench
Status: active
Retrieved at: 2026-08-09
Methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: SecCodeBench. Dataset type: public. Leaderboard date: 2026-02-16. Source ID: seccodebench. Evidence capture ID: curated-seccodebench-2026-08-09-ce147aec18e9. Source payload hash: ce147aec18e9354c4817af561e4326fc26c9aa5fb2fae3b3abce107c5ae614a5. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.

Source SecCodeBench

SecCodeBench-V2: Generation without hints (Show measurement, test conditions, and source)

Source value: 51.05
Score: 51.05
Metric: native generation pass@1
Unit: %
Benchmark version: 2.2.0
Category: cybersecurity
Task: generation
Task label: Generation without hints
Direction: Higher is better
Comparison cohort: seccodebench-v2-2-0:official:generation
Source type: official-leaderboard
Evaluator: SecCodeBench
Status: active
Retrieved at: 2026-08-09
Methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: SecCodeBench. Dataset type: public. Leaderboard date: 2026-02-16. Source ID: seccodebench. Evidence capture ID: curated-seccodebench-2026-08-09-ea0efe96af3b. Source payload hash: ea0efe96af3b9a3ff40cb16e820c61e6be6d652be27ce4f466af6ee5d583d804. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.

Source SecCodeBench

SecCodeBench-V2: Generation with security hints (Show measurement, test conditions, and source)

Source value: 70.65
Score: 70.65
Metric: security-aware generation pass@1
Unit: %
Benchmark version: 2.2.0
Category: cybersecurity
Task: generationHints
Task label: Generation with security hints
Direction: Higher is better
Comparison cohort: seccodebench-v2-2-0:official:generationHints
Source type: official-leaderboard
Evaluator: SecCodeBench
Status: active
Retrieved at: 2026-08-09
Methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: SecCodeBench. Dataset type: public. Leaderboard date: 2026-02-16. Source ID: seccodebench. Evidence capture ID: curated-seccodebench-2026-08-09-5a782a1bbabe. Source payload hash: 5a782a1bbabe9946ce4fb4771835246fcedde9c1186f456a2a364c1860c8531a. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.

Source SecCodeBench

SecCodeBench-V2: Weighted total (Show measurement, test conditions, and source)

Source value: 64.9
Score: 64.9
Metric: weighted pass@1
Unit: %
Benchmark version: 2.2.0
Category: cybersecurity
Task: total
Task label: Weighted total
Direction: Higher is better
Comparison cohort: seccodebench-v2-2-0:official:total
Source type: official-leaderboard
Evaluator: SecCodeBench
Status: active
Retrieved at: 2026-08-09
Methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.. Source model ID: curated/claude-opus-4.6. Source model name: Claude Opus 4.6. Provider reported by source: Anthropic. Evaluator: SecCodeBench. Dataset type: public. Leaderboard date: 2026-02-16. Source ID: seccodebench. Evidence capture ID: curated-seccodebench-2026-08-09-1a93d69f218e. Source payload hash: 1a93d69f218e7cddea5bc50ca7db1867e78031ce74a62a35f964a572e0a55048. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 98 project cases. Functional tests run before dynamic exploit checks and security scoring.

Source SecCodeBench

Page 1 of 3

Predecessor and successor

Predecessor: Claude Opus 4.5
Model lineage evidence
Anthropic writes "The new Claude Opus 4.6 improves on its predecessor's coding skills" and separately reports Opus 4.6 outperforming "its own predecessor (Claude Opus 4.5)" by 190 Elo points on GDPval-AA.
Anthropic "Introducing Claude Opus 4.6" announcement (2026-09-19)
Successor: Claude Opus 4.7
Model lineage evidence
Anthropic writes "Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks.".
Anthropic "Introducing Claude Opus 4.7" announcement (2026-09-19)

Sources and data date

Type	Evidence and data date
Speed	Anthropic, Non-reasoning, high. Output speed after the first streaming chunk, not total response time. Classified as deprecated by Artificial Analysis. Artificial Analysis maintains only the default 10K-input workload there. Retrieved September 8, 2026. Source: Artificial Analysis · Artificial Analysis (retrieved September 8, 2026)Speed measurement methodology
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · Context window data
Input	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Opus 4.6 model overview · Editorial description reviewed October 2, 2026
Output	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Opus 4.6 model overview · Editorial description reviewed October 2, 2026
Research status	Research date September 8, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	parameters-billions Original wording Anthropic does not publish parameter counts for this model.
Not found in the inspected sources	active-parameters-billions Original wording Anthropic does not publish active parameter counts for this model.
Additional source	Model metadata (retrieved September 8, 2026) · Anthropic, Introducing Claude Opus 4.6 · Editorial description reviewed October 2, 2026
Additional source	Scale AI (retrieved September 2, 2026)
Additional source	METR (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	SecCodeBench (retrieved August 9, 2026)
Additional source	Vals AI (retrieved September 29, 2026)
Additional source	Vals AI (retrieved September 29, 2026)
Additional source	Vals AI (retrieved September 29, 2026)
Additional source	Claude Opus 4.6 system card · Editorial description reviewed October 2, 2026
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Creative Commons Attribution 4.0
Additional source	Anthropic "Introducing Claude Opus 4.7" announcement (2026-09-19)

Claude Opus 4.6

Specifications and access

Published benchmarks

Humanity's Last Exam

Agent tasks

METR Time Horizon

MultiChallenge

PRBench Finance

PRBench Legal

VisualToolBench

Cybersecurity

SecCodeBench-V2

Recorded measurement details

Our tests

Pelican on a bicycle

Popularity

Published prices

Compare this model

Predecessor and successor

Sources and data date