Fortress ↗18.06%
Claude 4 Sonnet (thinking) · Scale Labs · Risk elicitation score ↗
Anthropic / MODEL EVIDENCE
Claude 4 Sonnet
4 benchmarks · 29 observations · 4 sources
Scores at a glance
4 benchmarks · large dot: Claude 4 Sonnet; small dots: closest models where reported
GAIA-2 ↗57.9%
Claude-4-Sonnet · GAIA-2 · Gaia2 full · Execution pass@1 · baseline ReAct · three runs ↗
15 more reported results
62.1% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Execution pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
59.8% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Search pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
60.6% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Search pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
24.2% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Ambiguity pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
27.3% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Ambiguity pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
38.1% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Adaptability pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
42.1% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Adaptability pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
8.1% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Time pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
8.5% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Time pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
27.7% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Noise pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
31.2% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Noise pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
27.9% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Agent2Agent pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
32.5% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Agent2Agent pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
34.8% · Claude-4-Sonnet · GAIA-2 · Gaia2 full · Overall pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
37.8% · Claude-4-Sonnet Thinking · GAIA-2 · Gaia2 full · Overall pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
Biomni-Eval ↗0.557 / 1
Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · mean task reward · Biomni-R0 · A1 scaffold and E1 environment ↗
10 more reported results
0.35 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.5 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.78 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.68 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.6 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.52 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.86 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.3 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · Patient gene detection · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.6 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · Rare disease diagnosis · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
0.38 / 1 · Claude 4 Sonnet · Biomni-Eval · Biomni-Eval1 · Screen gene retrieval · task reward · Biomni-R0 · A1 scaffold and E1 environment · Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
ALE-Bench ↗655.35 Performance points
claude-4-sonnet · Average contest performance ↗
Closest models by score
Similar positions on shared benchmark leaderboards. Smaller gaps are closer.
GPT-5 →OpenAI
3 of 4 shared benchmark comparisons
How similarity is measured
Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.
Sources
Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.
- Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments — published result table ↗
All 24 authors
Froger, Romain, Andrews, Pierre, Bettini, Matteo, Budhiraja, Amar, Cabral, Ricardo Silveira, Do, Virginie, Garreau, Emilien, Gaya, Jean-Baptiste, Laurençon, Hugo, Lecanu, Maxime, Malkan, Kunal, Mekala, Dheeraj, Ménard, Pierre, Bertran, Gerard Moreno-Torres, Piterbarg, Ulyana, Plekhanov, Mikhail, Rita, Mathieu, Rusakov, Andrey, Vorotilov, Vladislav, Wang, Mengjue, Yu, Ian, Benhalloum, Amine, Mialon, Grégoire, Scialom, Thomas
- Ale Bench — methodology and published results ↗
- Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
- Scale Labs — evaluation methodology and results ↗
Names used in source tables
Claude 4 Sonnet · Claude 4 Sonnet (thinking) · Claude-4-Sonnet · Claude-4-Sonnet Thinking · claude-4-sonnet