Rankings / Anthropic
Claude Sonnet 4
release date unknown
BenchAtlas Index
as of 2026-07-14
thinking
rank #284
4 families · 4 categories · medium
base configuration
rank #312
4 families · 4 categories · medium
Benchmark evidence
10 results
coding
| LiveCodeBench v6 problems=454 · window_start=2024-08-01 | thinking | 56.0 % | independent | LiveCodeBench Leaderboard |
| LiveCodeBench v6 problems=454 · window_start=2024-08-01 | 47.1 % | independent | LiveCodeBench Leaderboard |
factuality
| MASK | thinking | 95.3 % 93.0–97.6 | independent | Scale Labs |
| MASK | 89.3 % 87.3–91.3 | independent | Scale Labs |
knowledge science
| Humanity's Last Exam contamination=Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. · implementation=scale | thinking | 7.8 % 6.7–8.8 | independent | Scale Labs |
| Humanity's Last Exam contamination=Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. · implementation=scale | 5.5 % 4.6–6.4 | independent | Scale Labs |
multimodal
| VISTA | thinking | 45.5 % 45.3–45.7 | independent | Scale Labs |
| VISTA | 43.2 % 42.7–43.7 | independent | Scale Labs |
reasoning math
| EnigmaEval contamination=Potential contamination warning: This model was evaluated after the public release of EnigmaEval, allowing model builder access to the prompts and solutions. | thinking | 3.1 % 2.1–4.1 | independent | Scale Labs |
| EnigmaEval contamination=Potential contamination warning: This model was evaluated after the public release of EnigmaEval, allowing model builder access to the prompts and solutions. | 2.2 % 1.4–3.0 | independent | Scale Labs |
Agent + model results
systems, not bare-model scores
| agent + model ante + Claude Sonnet 4 | Terminal-Bench 1.0 | 54.8 % | unverified | Terminal-Bench Leaderboard |
| agent + model Droid + Claude Sonnet 4 | Terminal-Bench 1.0 | 50.5 % | unverified | Terminal-Bench Leaderboard |
| agent + model chaterm + Claude Sonnet 4 | Terminal-Bench 1.0 | 49.3 % | unverified | Terminal-Bench Leaderboard |
| agent + model engine-labs + Claude Sonnet 4 | Terminal-Bench 1.0 | 44.8 % | unverified | Terminal-Bench Leaderboard |
| agent + model letta + Claude Sonnet 4 | Terminal-Bench 1.0 | 42.5 % | unverified | Terminal-Bench Leaderboard |
| agent + model OpenHands + Claude Sonnet 4 | Terminal-Bench 1.0 | 41.3 % | unverified | Terminal-Bench Leaderboard |
| agent + model goose + Claude Sonnet 4 | Terminal-Bench 1.0 | 41.3 % | unverified | Terminal-Bench Leaderboard |
| agent + model orchestrator + Claude Sonnet 4 | Terminal-Bench 1.0 | 37.0 % | unverified | Terminal-Bench Leaderboard |
| agent + model terminus-2 + Claude Sonnet 4 | Terminal-Bench 1.0 | 36.4 % | community | Terminal-Bench Leaderboard |
| agent + model Claude Code + Claude Sonnet 4 | Terminal-Bench 1.0 | 35.5 % | unverified | Terminal-Bench Leaderboard |
| agent + model goose + Claude Sonnet 4 | Terminal-Bench 1.0 | 34.3 % | unverified | Terminal-Bench Leaderboard |
| agent + model mini-SWE-agent + Claude Sonnet 4 | Terminal-Bench 1.0 | 12.8 % | community | Terminal-Bench Leaderboard |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.