BlogLLM Benchmark Scores as Published, Side by Side
Claude Opus 4.6: Benchmark Scores as Published
Every result the catalog holds for Claude Opus 4.6: 76 tests in 16 areas, from 15 pages. 23 printed by Anthropic itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer AnthropicReleased 2026-02-05Context window 1MMax output 128KInput Text, imageOutput TextWeights ClosedData version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| APEX-Agents | 29.8% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| JobBench | 36.6 | Alibaba, Qwen3.8-Flash-Next model cardchecked 2026-10-15 | Effort: max |
| GDPval-AA (Elo) | 1606 | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 3 more pages | - |
| Vals Finance Agent v1.1 | 60.1% | Anthropic, Claude Opus 4.7checked 2026-10-15; printed on 1 more page | - |
| Vending-Bench 2 (final balance in USD) | $8017.59 | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| CoWorkBench | 68.2 | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| Vals Finance Agent (version not stated) | 60.7% | Anthropic, Claude Opus 4.6checked 2026-10-15 | - |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-bench Verified | 80.8% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 7 more pages | - |
| SWE-Bench Pro | 53.4% | Anthropic, Claude Opus 4.7checked 2026-10-15; printed on 7 more pages | - |
| Terminal-Bench 2.1 | 78.2% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15 | Effort: max; Harness: Terminus |
| SWE-bench Multilingual | 77.5% | Alibaba, Qwen3.8-Flash-Next model cardchecked 2026-10-15; printed on 2 more pages | Effort: max; Harness: mini-SWE-agent; Context: 256000 |
| Terminal-Bench 2.0 | 65.4% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 7 more pages | Harness: Terminus-2 (reproduced by Anthropic) |
| NL2Repo-Bench | 47.6 | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 2 more pages | Effort: max; Harness: Claude Code |
| Multi-SWE-bench | 50.3% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| QwenSWEBench | 63.8 | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15 | Effort: max; Harness: Claude Code |
| VIBE-Pro | 55.6% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| SWE-MM | 27.1% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15 | Effort: max; Harness: Claude Code |
ML Engineering
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MLE-bench Lite (medal rate) | 75.7% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPQA Diamond | 91.3% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 10 more pages | - |
| AIME 2025 | 99.8% | Microsoft, Microsoft AI: Introducing MAI-Thinking-1checked 2026-10-19; printed on 1 more page | - |
| SciCode | 52% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
| IMO-AnswerBench | 75.3% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
| AIME 2026 | 96.7% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| HMMT November 2025 | 96.3% | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| HMMT February 2026 | 96.2% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
| MathArena Apex Shortlist | 85.9% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| MathArena Apex | 34.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AA-LCR | 71% | MiniMax, MiniMax-M2.5 model cardchecked 2026-10-15 | - |
| MRCR v2 8-needle 128K | 84% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: max; Context: 128k (average) |
| CorpusQA 1M | 71.7% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M |
| MRCR 1M (MMR) | 92.9% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M |
Computer Use
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| OSWorld-Verified | 72.7% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 4 more pages | - |
| AndroidWorld | 62% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
Multimodal Understanding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CharXiv Reasoning (no tools) | 69.1% | Anthropic, Claude Opus 4.7checked 2026-10-15; printed on 3 more pages | Tools: none |
| MMMU-Pro | 73.9% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 3 more pages | Tools: none |
| MathVision | 65.5% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 2 more pages | Effort: max; Tools: none |
| CharXiv Reasoning (with tools) | 84.7% | Anthropic, Claude Opus 4.7checked 2026-10-15; printed on 1 more page | Tools: enabled; exact tools in source |
| BabyVision (with tools) | 38.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max; Tools: python |
| MMMU-Pro (with tools) | 77.3% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 2 more pages | Tools: enabled; exact tools in source |
| LVBench | 63% | Alibaba, Qwen3.8-Flash-Next model cardchecked 2026-10-15 | Effort: max |
| OmniDocBench 1.5 | 86.6% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15 | Effort: max |
| BabyVision (no tools) | 12.6% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max; Tools: none |
| ERQA | 40.8% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| RealWorldQA | 73.9% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| MathVision (with tools) | 84.6% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max; Tools: python |
| V* Bench (with tools) | 86.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max; Tools: python |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 73.8% | Anthropic, Claude Opus 4.7checked 2026-10-15; printed on 1 more page | - |
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 40% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 9 more pages | Effort: max; Tools: none |
| MMLU-Pro | 89.1% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| Humanity's Last Exam (with tools) | 53% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 6 more pages | Effort: max; Tools: enabled; exact tools in source |
| SimpleQA Verified | 46.2% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| ARC-AGI-2 | 68.8% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 2 more pages | Effort: max |
| Chinese SimpleQA | 76.4% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
Code Generation
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| LiveCodeBench v6 | 88.8% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
| LiveCodeBench (version not stated) | 88.8% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max |
| OJBench (Python) | 60.3% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max |
Tool Use and Workflows
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Toolathlon | 47.2% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 3 more pages | Effort: max |
| τ²-bench Telecom | 99.3% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 2 more pages | - |
| MCP Atlas | 59.5% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 3 more pages | Effort: max |
| 62.7% | Anthropic, Claude Opus 4.6checked 2026-10-15 | Effort: high | |
| τ²-bench Retail | 91.9% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 2 more pages | - |
| MCP Atlas (public set) | 73.8% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| MCPMark | 56.7% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max |
| Claw-Eval pass^3 | 70.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max |
| τ³-bench (domain not stated) | 72.4% | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| MM-ClawBench | 75.4% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| ClawEval-MM average score | 54.7 | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| ClawEval-MM pass@3 | 52.5% | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| Claw-Eval pass@3 | 82.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max |
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| BrowseComp | 86.8% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 6 more pages | Effort: max; Harness: multi-agent harness; Tools: web search, web fetch, programmatic tool calling |
| 84% | Anthropic, Claude Opus 4.6checked 2026-10-15 | Effort: max; Tools: web search, web fetch, programmatic tool calling | |
| DeepSearchQA (accuracy) | 80.6% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max; Tools: search, code interpreter, browsing |
| DeepSearchQA (F1) | 91.3% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: max; Tools: search, code interpreter, browsing |
Instruction Following
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| IFBench | 62.5 | Alibaba, Qwen3.8-27B model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
Multilingual
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MMMLU | 91.1% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 3 more pages | - |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Self-preference bias, Claude-identity system prompt (0-9 scale)lower is better | 0.2 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
| Self-preference bias, no system prompt (0-9 scale)lower is better | 0.2 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
Preference and Composite Rankings
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 53 | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked. Cost per task is measured by Vals AI.
| Price | Value | Printed by | Setup as printed |
|---|---|---|---|
| Input | $5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cached input | $0.5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cache write | $6.25 | Anthropic, Anthropic API pricingchecked 2026-10-15 | Cache: 5 minute cache |
| Cache write, 1 hour | $10 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Output | $25 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Input, batch | $2.5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Output, batch | $12.5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cost per taskVals AI, USD, lower is better | $8.69 | Vals AI, Vibe Code Bench v1.1 leaderboardchecked 2026-10-27 | Harness: OpenHands |
More from Anthropic
- Claude Haiku 5.5
- Claude Sonnet 5.5
- Claude Opus 5.5
- Claude Fable 5.1
- Claude Mythos 5.1
- Claude Opus 5
- Claude Sonnet 5
- Claude Fable 5
- Claude Mythos 5
- Claude Opus 4.8
- Claude Opus 4.7
- Claude Mythos Preview
- Claude Sonnet 4.6
- Claude Opus 4.5
- Claude Haiku 4.5
- Claude Sonnet 4.5
- Claude Opus 4.1
- Claude Opus 4
- Claude Sonnet 4
- Claude 3.7 Sonnet
- Claude 3.5 Sonnet
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.