List your company

Tool-calling on ACEBench English

25Multistep (Agent)Qwen3-4B-RL (ours)
9.413.4517.521.55May 28, 2026

Evaluation results

MethodLinks
Qwen3-4B-RL (ours)Source=Open-source2026.0525309079.580708072909277.5
Nova-1-ProSource=Closed-source2026.052056.794.381.588.5747662448481.4
Nova-1-LiteSource=Closed-source2026.052060866078.5687414467873.4
Gemma3-27B-InstructSource=Open-source2026.052056.788.35581667670329076.8
Magistral-Small-24B-2509Source=Open-source2026.052050915882707440184867.4
Claude-4-SonnetSource=Closed-source2026.051553.3947986.5768696769481.8
Qwen3-4B (base)Source=Open-source2026.05102077.364.766.5647658847465.4