Victor Barres et al.

τ²-Bench Tool-Agent-User Evaluation

In index
Agentic
Best in agentic
Source ↗

Models150

Top result99.1%GLM-5.2

Top-3 spread0.6 pts

Year2025

Result and price

90 of 150 results

Results

150 results

#ModelResult

01GLM-5.299.1%

02GPT-5.498.9%

03Claude Fable 598.5%

03Step 3.7 Flash98.5%

03GLM-5-Turbo98.5%

03GLM-5V-TurboAbove fit98.5%

07GLM-598.2%

08GPT-5.598.0%

09GLM-5.197.7%

09Qwen3.6 Plus97.7%

09Grok 4.397.7%

12DeepSeek V4 Pro 081396.2%

13Kimi K2.695.9%

13Qwen 3.6 Max (preview)95.9%

13Kimi K2.5 (Reasoning)95.9%

13Kimi K2.595.9%

13GLM-4.795.9%

18Gemini 3.1 Pro95.6%

18Qwen3.5 397B (Reasoning)95.6%

18Qwen3.5 397B95.6%

21Gemini 3.5 Flash95.3%

21Qwen3.6-35B-A3B95.3%

23MiMo-V2-Pro95.0%

24Qwen3.7 Max94.7%

25Claude Opus 4.894.4%

26MiMo-V2.5-Pro94.2%

26Qwen3.6-27B94.2%

26Mistral Medium 3.5 128B94.2%

29Qwen3.5-27B93.9%

30Qwen3.5-122B-A10B93.6%

31GPT-5.4 mini93.4%

32Grok 4.1 Fast (Reasoning)93.3%

33Qwen3.7 Plus93.0%

34GPT-5.4 nano92.5%

35Claude Opus 4.6 (Adaptive)92.1%

35GPT-5.2-Codex92.1%

35Qwen3.5-4BAbove fit92.1%

38Muse Spark91.5%

39MiMo-V2-Omni91.2%

40Kimi K2.7 Code90.1%

40Trinity-Large-ThinkingAbove fit90.1%

40Trinity-Large-PreviewAbove fit90.1%

43Claude Opus 4.5 Thinking89.5%

44Qwen3.5-35B-A3B89.2%

45MiniMax M388.9%

46Claude Opus 4.7 (Adaptive)88.6%

47LFM2.5-8B-A1BAbove fit88.1%

48Gemini 3 Pro87.1%

49Spark-X2.6-4B86.9%

50GPT-5 (medium)86.5%

51GPT-5.6 Terra86.3%

51Claude Opus 4.586.3%

51Solar Pro 3Above fit86.3%

54GPT-5.3 Codex86.0%

54Ling 2.6 FlashAbove fit86.0%

56GPT-5.6 Sol85.1%

57Command A+Above fit85.0%

58GPT-5.284.8%

58Claude Opus 4.684.8%

58GPT-5 (high)84.8%

58MiniMax M2.784.8%

62MiMo-V2-Flash83.9%

63Nemotron 3 Ultra83.3%

64GPT-5.1-Codex-Max83.0%

64GPT-5.1-Codex83.0%

66GPT-5.181.9%

67o380.7%

68LLaDA2.2-flash80.3%

69Ternary Bonsai 2 27B80.2%

70Claude Sonnet 4.679.5%

71DeepSeek V3.278.9%

72Agents-A1-4B78.2%

73GLM-4.676.9%

74Grok Code Fast 175.7%

75Grok 474.9%

76Qwen3 Max74.3%

76K-Exaone74.3%

78Claude Opus 4.774.0%

79Claude 4 Opus Thinking73.4%

80Claude 4.1 Opus Thinking71.4%

81Grok 4 Fast (Reasoning)65.8%

81GPT-OSS 120B65.8%

83Grok 4.1 Fast63.7%

84o162.6%

85Kimi K261.1%

86GPT-OSS 20B60.2%

87Gemma 4 31B59.9%

88LLaDA2.2-miniAbove fit57.5%

89o4-mini55.6%

90Nemotron 3.5 Bolt E4B BF1655.1%

91Claude 3.7 Sonnet Thinking54.7%

92Gemini 2.5 Pro54.1%

93GPT-4.1 mini52.9%

94Nemotron 3.5 Bolt E4B52.3%

94Claude 4 Sonnet52.3%

96Claude 3.7 Sonnet50.0%

97GPT-4.147.1%

97DeepSeek V3 032447.1%

99Sarvam 105B46.8%

100GLM-4.5-Air46.5%

101Nemotron 3 Nano Omni 30B A3B45.3%

102Gemma 4 26B A4B43.6%

103Gemini 3 Flash43.3%

104Mistral Small 441.2%

—Mistral Small 4 (Reasoning)41.2%

105DeepSeek V3.1 (Reasoning)37.4%

105North Mini Code37.4%

107DeepSeek-R136.5%

108Gemma 4 12B36.3%

109DeepSeek V3.134.8%

110Sarvam 30B34.5%

111Solar Pro 231.9%

112Mistral Large 230.7%

113Gemini 2.0 Flash29.5%

114o3-mini28.7%

115Devstral Small28.4%

116Magistral Small26.6%

116Llama 3.3 70B26.6%

116Ultravox v0.6 Llama 3.3 70B26.6%

119Nemotron 3 Nano 30B25.4%

120GPT-4o25.1%

120Mistral Small 3.125.1%

122Mistral Large 324.6%

122Claude 3.5 Haiku24.6%

124Mistral Medium 324.3%

125Magistral Medium23.1%

126DeepSeek V322.8%

127Llama 3.1 70B21.6%

128Qwen3-Omni-30B-A3B-Thinking21.3%

129Claude 3 Haiku21.1%

130Gemma 4 E4B20.8%

130Gemma 4 E2B20.8%

132Exaone 4.0 1.2B20.5%

133Granite-4.0-H-1B19.6%

134Gemini 2.5 Flash-Lite19.0%

134Llama 3.1 405B19.0%

136Llama 4 Maverick17.8%

137GPT-4.1 nano17.3%

138Qwen3-Omni-30B-A3B-Instruct16.4%

139Llama 4 Scout15.5%

140Gemini 2.5 Flash14.9%

141Granite-4.0-H-350M14.6%

142Nova Pro14.0%

143Granite-4.0-350M13.2%

144Nemotron Ultra 253B11.4%

145Gemma 3 27B10.5%

146LFM2.5-VL-1.6B-Extract8.5%

147Exaone 4.0 32B4.1%

148Phi-4-reasoning0.0%

148Phi-40.0%