BlogLLM Benchmark Scores as Published, Side by Side
Claude Sonnet 4.6: Benchmark Scores as Published
Every result the catalog holds for Claude Sonnet 4.6: 63 tests in 17 areas, from 12 pages. 21 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 2026-02-17Context window 1MMax output 128KInput Text, imageOutput TextWeights ClosedData version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GDPval-AA v2 | 1395 | Anthropic, Claude Sonnet 5checked 2026-10-15 | - |
| APEX-Agents | 26.2% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| GDPval-AA (Elo) | 1633 | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | - |
| Vals Finance Agent v2 | 51% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | - |
| Vals Finance Agent v1.1 | 63.3% | Anthropic, Claude Sonnet 4.6checked 2026-10-15 | - |
| GDPval rubrics (pointwise rubric score) | 75.65% | MiniMax, MiniMax M3checked 2026-10-15 | - |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-bench Verified | 79.6% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 4 more pages | - |
| SWE-Bench Pro | 58.1% | Anthropic, Claude Sonnet 5checked 2026-10-15; printed on 1 more page | - |
| Terminal-Bench 2.1 | 67% | Anthropic, Claude Sonnet 5checked 2026-10-15 | - |
| Terminal-Bench 2.0 | 59.1% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | Effort: none; Harness: Terminus-2 (reproduced by Anthropic) |
| SWE-Atlas-QnA (setting not stated) | 31.2% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| Multi-SWE-bench | 51% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| SWE-Atlas-Test Writing | 31.76% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| VIBE-Pro | 56.1% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
ML Engineering
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MLE-bench Lite (medal rate) | 72.7% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPQA Diamond | 89.9% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | - |
| AIME 2025 | 95.6% | Microsoft, Microsoft AI: Introducing MAI-Thinking-1checked 2026-10-19; printed on 1 more page | - |
| SciCode | 47% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: max |
| BeyondAIME | 47.3% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| LongBench v2 | 66% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| MRCR v2 8-needle 128K | 84.9% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 1 more page | Effort: max; Context: 128k (average) |
| CorpusQA (AI-judge score)AI-judge score | 79 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| GraphWalks up to 128k (overall F1) (GraphWalks overall F1, up to 128k)GraphWalks overall F1, up to 128k | 96% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Computer Use
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| OSWorld-Verified | 72.5% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 3 more pages | - |
Multimodal Understanding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CharXiv Reasoning (no tools) | 72.4% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | Tools: none |
| MMMU-Pro | 74.5% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 3 more pages | Tools: none |
| MMMU-Pro (with tools) | 75.6% | Anthropic, Claude Sonnet 4.6checked 2026-10-15 | Tools: enabled; exact tools in source |
| OmniDocBench 1.5 | 86.9% | MiniMax, MiniMax M3checked 2026-10-15 | - |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberSecEval 4 Autocomplete (score)score | 56 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| CyberSecEval 4 Instruct (score)score | 62 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 33.2% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 3 more pages | Effort: max; Tools: none |
| MMLU-Pro | 87% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| Humanity's Last Exam (with tools) | 49% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | Effort: max; Tools: enabled; exact tools in source |
| SimpleQA Verified | 29% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| ARC-AGI-2 | 60.4% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | Effort: high |
| 58.3% | Anthropic, Claude Sonnet 4.6checked 2026-10-15 | Effort: max | |
| LongFact claim precision (LongFact claim-level precision)LongFact claim-level precision | 98% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| TruthfulQA (accuracy)accuracy | 88% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Tool Use and Workflows
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Toolathlon | 44.8% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| τ³-bench Banking | 28.4% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max; Harness: gpt-5.2 user simulator (low), 4 trials; terminal- or embedding-based agentic search retrieval, highest score |
| τ²-bench Telecom | 97.9% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 1 more page | - |
| MCP Atlas | 61.3% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | - |
| τ²-bench Retail | 91.7% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 1 more page | - |
| MCP Atlas (public set) | 61.3% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| Claw-Eval pass^3 | 68.3% | MiniMax, MiniMax M3checked 2026-10-15 | - |
| τ³-bench Airline | 83% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max; Harness: gpt-5.2 user simulator (low), 4 trials; Sierra-reported for Claude Sonnet 4.5 and Qwen3.5 |
| τ³-bench Retail | 75.9% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max; Harness: gpt-5.2 user simulator (low), 4 trials; Sierra-reported for Claude Sonnet 4.5 and Qwen3.5 |
| τ³-bench Telecom | 70.4% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max; Harness: gpt-5.2 user simulator (low), 4 trials; Sierra-reported for Claude Sonnet 4.5 and Qwen3.5 |
| MM-ClawBench | 64.2% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| BFCL v3 (accuracy)accuracy | 76% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| BrowseComp | 74.7% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 3 more pages | Effort: max; Tools: web search, web fetch, programmatic tool calling |
| DRACO | 75.8% | MiniMax, MiniMax M3checked 2026-10-15 | - |
Instruction Following
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| IFBench | 50 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19; printed on 1 more page | Effort: max |
| MultiChallenge | 57% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| COLLIE | 67.7% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| AdvancedIF (rubric-level score)rubric-level score | 86 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Multilingual
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MMMLU | 89.3% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 1 more page | - |
Healthcare
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| HealthBench Professional | 38% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
| MedXpertQA Text (accuracy)accuracy | 49% | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Self-preference bias, Claude-identity system prompt (0-9 scale)lower is better | 0.31 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
| Self-preference bias, no system prompt (0-9 scale)lower is better | 0.17 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
| AIR-Bench 2024 (aggregate score)aggregate score | 88 | Microsoft, Microsoft AI: MAI-Thinking-1 technical reportchecked 2026-10-19 | Effort: max |
Engineering and Design
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Blueprint-Bench 2 (normalized score) | 6.7% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | - |
Preference and Composite Rankings
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 52 | 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 | $3 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cached input | $0.3 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cache write | $3.75 | Anthropic, Anthropic API pricingchecked 2026-10-15 | Cache: 5 minute cache |
| Cache write, 1 hour | $6 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Output | $15 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Input, batch | $1.5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Output, batch | $7.5 | Anthropic, Anthropic API pricingchecked 2026-10-15 | - |
| Cost per taskVals AI, USD, lower is better | $5.91 | 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 Opus 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.