Rankings / Unknown
Claude 4 Sonnet
released 2025-05-22
BenchAtlas Index
as of 2026-07-14
thinking
rank #273
11 families · 5 categories · high
no reasoning
rank #342
10 families · 5 categories · high
Benchmark evidence
29 results
agentic coding
| τ²-Bench | thinking | 64.6 % | independent | Artificial Analysis |
| τ²-Bench | no reasoning | 52.3 % | independent | Artificial Analysis |
| τ²-Bench subset=banking | thinking | 13.8 % | independent | Artificial Analysis |
coding
| LiveCodeBench v6 implementation=artificial-analysis | thinking | 65.5 % | independent | Artificial Analysis |
| LiveCodeBench v6 implementation=artificial-analysis | no reasoning | 44.9 % | independent | Artificial Analysis |
| SciCode | thinking | 40.0 % | independent | Artificial Analysis |
| SciCode | no reasoning | 37.3 % | independent | Artificial Analysis |
external indices
| AA Coding Index | thinking | 37.6 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | thinking | 28.9 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | no reasoning | 25.5 points | independent | Artificial Analysis |
| AA Math Index | thinking | 74.3 points | independent | Artificial Analysis |
| AA Math Index | no reasoning | 38.0 points | independent | Artificial Analysis |
knowledge science
| GPQA Diamond implementation=artificial-analysis | thinking | 77.7 % | independent | Artificial Analysis |
| GPQA Diamond implementation=artificial-analysis | no reasoning | 68.3 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | thinking | 9.6 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | no reasoning | 4.0 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | thinking | 84.2 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | no reasoning | 83.7 % | independent | Artificial Analysis |
long context instruction
| AA-LCR | thinking | 64.7 % | independent | Artificial Analysis |
| AA-LCR | no reasoning | 44.3 % | independent | Artificial Analysis |
| IFBench | thinking | 54.7 % | independent | Artificial Analysis |
| IFBench | no reasoning | 45.4 % | independent | Artificial Analysis |
| MultiChallenge | thinking | 57.1 % 56.3–57.9 | independent | Scale Labs |
reasoning math
| AIME implementation=artificial-analysis | thinking | 77.3 % | independent | Artificial Analysis |
| AIME year=2025 · implementation=artificial-analysis | thinking | 74.3 % | independent | Artificial Analysis |
| AIME implementation=artificial-analysis | no reasoning | 40.7 % | independent | Artificial Analysis |
| AIME year=2025 · implementation=artificial-analysis | no reasoning | 38.0 % | independent | Artificial Analysis |
| MATH-500 implementation=artificial-analysis | thinking | 99.1 % | independent | Artificial Analysis |
| MATH-500 implementation=artificial-analysis | no reasoning | 93.4 % | independent | Artificial Analysis |
Agent + model results
systems, not bare-model scores
| agent + model Refact.ai Agent + Claude 4 Sonnet | SWE-bench Multimodal | 35.6 % | unverified | SWE-bench Leaderboard |
| agent + model Artificial Analysis harness + Claude 4 Sonnet | Terminal-Bench 2.1 | 36.3 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + Claude 4 Sonnet | Terminal-Bench Hard | 31.1 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + Claude 4 Sonnet | Terminal-Bench Hard | 27.3 % | independent | Artificial Analysis |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.