Rankings / Anthropic
Claude 4.5 Sonnet
released 2025-09-29
BenchAtlas Index
as of 2026-07-14
thinking
rank #174
9 families · 5 categories · high
no reasoning
rank #249
9 families · 5 categories · high
Benchmark evidence
24 results
agentic coding
| τ²-Bench | thinking | 78.1 % | independent | Artificial Analysis |
| τ²-Bench | no reasoning | 70.5 % | independent | Artificial Analysis |
| τ²-Bench subset=banking | thinking | 19.0 % | independent | Artificial Analysis |
coding
| LiveCodeBench v6 implementation=artificial-analysis | thinking | 71.4 % | independent | Artificial Analysis |
| LiveCodeBench v6 implementation=artificial-analysis | no reasoning | 59.0 % | independent | Artificial Analysis |
| SciCode | thinking | 44.7 % | independent | Artificial Analysis |
| SciCode | no reasoning | 42.8 % | independent | Artificial Analysis |
external indices
| AA Coding Index | thinking | 52.1 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | thinking | 36.4 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | no reasoning | 29.3 points | independent | Artificial Analysis |
| AA Math Index | thinking | 88.0 points | independent | Artificial Analysis |
| AA Math Index | no reasoning | 37.0 points | independent | Artificial Analysis |
knowledge science
| GPQA Diamond implementation=artificial-analysis | thinking | 83.4 % | independent | Artificial Analysis |
| GPQA Diamond implementation=artificial-analysis | no reasoning | 72.7 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | thinking | 17.3 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | no reasoning | 7.1 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | thinking | 87.5 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | no reasoning | 86.0 % | independent | Artificial Analysis |
long context instruction
| AA-LCR | thinking | 65.7 % | independent | Artificial Analysis |
| AA-LCR | no reasoning | 51.3 % | independent | Artificial Analysis |
| IFBench | thinking | 57.3 % | independent | Artificial Analysis |
| IFBench | no reasoning | 42.6 % | independent | Artificial Analysis |
reasoning math
| AIME year=2025 · implementation=artificial-analysis | thinking | 88.0 % | independent | Artificial Analysis |
| AIME year=2025 · implementation=artificial-analysis | no reasoning | 37.0 % | independent | Artificial Analysis |
Agent + model results
systems, not bare-model scores
| agent + model apex2 + Claude 4.5 Sonnet | Terminal-Bench 1.0 | 64.5 % | unverified | Terminal-Bench Leaderboard |
| agent + model chaterm + Claude 4.5 Sonnet | Terminal-Bench 1.0 | 63.8 % | unverified | Terminal-Bench Leaderboard |
| agent + model Artificial Analysis harness + Claude 4.5 Sonnet | Terminal-Bench 2.1 | 55.8 % | independent | Artificial Analysis |
| agent + model final + Claude 4.5 Sonnet | Terminal-Bench 2.0 | 42.7 % | unverified | Terminal-Bench Leaderboard |
| agent + model Artificial Analysis harness + Claude 4.5 Sonnet | Terminal-Bench Hard | 35.6 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + Claude 4.5 Sonnet | Terminal-Bench Hard | 28.8 % | independent | Artificial Analysis |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.