Vals AI

LegalBench

Evaluating language models on a wide range of open source legal reasoning tasks.

Not in index
Legal
Source ↗

Models139

Top result88.6%Claude Fable 5

Top-3 spread1.2 pts

Updated21 Sept 2026

Result and cost

Results

140 results

#ModelResultCostTime

Claude Fable 588.6%$0.0319.0 s

Claude Fable 5.188.5%$0.05212 s

Gemini 3.1 Pro Preview87.4%<$0.0110 s

Gemini 3.7 Flash87.3%<$0.012.2 s

Gemini 3 Pro87.0%$0.018.3 s

Gemini 3.8 Flash87.0%<$0.013.3 s

Claude Opus 587.0%$0.0124.4 s

GPT-5.6 Sol87.0%$0.0136.2 s

Gemini 3 Flash Preview86.9%<$0.014.7 s

Gemini 3.6 Flash86.7%<$0.013.0 s

GPT-5.586.5%$0.01518 s

Grok 4.686.3%$0.01527 s

GPT-5.486.0%$0.02328 s

GPT-586.0%—19 s

Grok 4.586.0%<$0.0168 s

GPT-5.185.7%<$0.017.0 s

MiniMax M385.4%<$0.018.2 s

Claude Opus 4.6 (Adaptive)85.3%<$0.013.8 s

Muse Spark 1.285.3%<$0.0130 s

Claude Opus 4.785.3%<$0.0127 s

GPT-5.6 Terra85.1%<$0.012.8 s

Qwen3.5 Plus Thinking85.1%<$0.0130 s

Muse Spark 1.185.0%<$0.0117 s

Qwen3.7 Max84.9%<$0.0112 s

GLM-5.384.8%<$0.0122 s

Claude Opus 4.5 Thinking84.6%$0.0377.6 s

Grok 4.384.5%<$0.016 min

GLM-5.184.4%<$0.0114 s

Grok 4.784.4%$0.01126 s

Gemini 2.5 Pro Exp 03.2584.3%—4.0 s

Qwen3.5 Flash84.3%<$0.0117 s

Qwen3.6 Plus84.2%<$0.0119 s

Muse Spark84.2%<$0.0134 s

Claude Sonnet 4.5 Thinking84.1%—13 s

GLM-5.284.1%<$0.015.8 s

Gemini 3.5 Flash-Lite84.1%<$0.011.7 s

GLM-5 Thinking84.1%<$0.0116 s

GPT-5.6 Luna84.0%<$0.015.9 s

MiniMax M2.784.0%<$0.018.9 s

GLM-5.3-Flash83.9%<$0.0113 s

Claude Sonnet 583.9%<$0.014.9 s

Gemini 2.5 Flash Preview 04.1783.8%—0.5 s

Gemini 3.1 Flash Lite Preview83.8%<$0.012.7 s

o383.8%—6.0 s

45TEHy4 preview83.8%$0.0132 min

Qwen3.8 Max83.6%$0.0136 s

Gemini 3.5 Flash83.6%<$0.0135 s

Claude Opus 4.883.6%<$0.013.1 s

Claude Opus 4.183.5%—2.8 s

GLM-4.783.4%<$0.0131 s

DeepSeek V4.1 Flash83.3%<$0.016.5 s

Grok 483.2%—27 s

Grok 3 Mini Fast High Reasoning83.1%—5.1 s

GPT-4.183.1%—0.5 s

Claude Opus 483.1%—3.7 s

56INMercury 2.583.1%<$0.013.9 s

57TMInkling-Small83.0%<$0.0118 s

Claude Sonnet 483.0%—4.8 s

59TMInkling82.9%$0.01323 s

Claude Opus 4.582.8%$0.0132.4 s

GPT-5.282.8%$0.01817 s

Gemini 2.5 Flash Preview Thinking82.6%—2.0 s

Grok 3 [Beta]82.6%—0.7 s

Claude 3.7 Sonnet Thinking82.5%—6.5 s

Gemini 2.5 Flash (Preview)82.5%—1.2 s

Grok 4.1 Fast (Reasoning)82.5%<$0.018.6 s

Qwen3.8-27B82.4%<$0.0148 s

Claude Sonnet 4.682.1%<$0.0122 s

Claude Sonnet 4 Thinking82.1%—11 s

Gemini 2.5 Flash Lite Preview Thinking82.0%—3.1 s

Grok 3 Mini Fast Low Reasoning81.9%—3.4 s

Qwen3 Maxqwen3-max81.9%—0.0 s

GPT-5 mini81.8%—9.4 s

74MAKimi K2 Instruct81.5%—1.2 s

DeepSeek V4 Pro 042381.3%<$0.0120 s

Claude Haiku 4.5 Thinking81.2%—10 s

GPT-4o81.2%—0.5 s

Grok 4 Fast (Reasoning)80.6%—3.8 s

GPT-4 Turbo80.5%—1.4 s

o180.4%—8.3 s

Qwen3 Maxqwen3-max-preview80.3%—1.9 s

82MAKimi K2 Thinking80.2%<$0.0144 s

Qwen3 235B A22B80.2%—12 s

Claude 3.7 Sonnet80.0%—1.2 s

MiniMax M2.580.0%<$0.0116 s

86COCommand A79.7%—0.8 s

87INLing 3.0 Flash79.7%<$0.013.9 s

GLM-4.679.6%—30 s

Qwen2.5 72B Instruct Turbo79.4%—0.8 s

DeepSeek V379.2%—4.4 s

o4-mini (high)79.2%—4.0 s

Mistral Large79.1%<$0.011.7 s

Grok 4.1 Fast Non Reasoning79.1%<$0.011.8 s

Gemini 2.5 Flash Lite Preview79.0%—0.5 s

MiMo-V2.578.9%<$0.0111 s

Grok 4 Fast Non Reasoning78.4%—0.9 s

Gemini 2.0 Flash 00178.4%—0.4 s

GPT-4.1 mini78.0%—0.5 s

GPT-5.4 nano77.9%<$0.012.7 s

100

Llama4 Maverick Instruct Basic77.8%—0.5 s

101

Grok 4.20 (Reasoning)77.7%<$0.017.6 s

102

DeepSeek V4 Flash 042377.7%<$0.013.9 s

103

Llama 3.3 70B Instruct Turbo77.2%—0.5 s

104

MiMo-V2.5-Pro77.1%<$0.0127 s

105

DeepSeek V3.2 (Thinking)76.1%<$0.0131 s

106

GPT-OSS 120B75.9%—7.9 s

107

GLM-4.575.6%—16 s

108

MiniMax M2.175.4%<$0.0111 s

109POLaguna M.175.1%—42 s

110ALJamba 1.5 Large74.2%—0.8 s

111

Llama 4 Scout 17B 16E Instruct72.0%—0.4 s

112ALJamba Large 1.672.0%—0.8 s

113

o3-mini71.5%—12 s

114POLaguna XS.271.0%—12 s

115

GPT-OSS 20B70.8%—7.2 s

116

Claude 3.5 Haiku70.3%—1.1 s

117

Gemini 1.0 Pro 00270.2%—0.5 s

118ALJamba Mini 1.669.7%—0.3 s

119

Qwen2.5 7B Instruct Turbo69.6%—0.4 s

120

Mistral Small69.2%—0.6 s

121

Gemini 1.5 Pro 00269.1%—0.9 s

122COCommand R Plus68.3%—0.4 s

123

Gemma 2 9B IT68.1%—0.6 s

124

Gemma 2 27B68.0%—0.9 s

125

DeepSeek-R167.3%—19 s

126ALJamba 1.5 Mini66.6%—0.3 s

127

DeepSeek V3.266.0%<$0.012.2 s

128

GPT-3.5 Turbo64.4%—0.5 s

129

Gemini 1.5 Flash 00263.2%—0.5 s

130

Llama 2 70B Hf62.4%—14 s

131

Mistral Medium62.0%—0.7 s

132

GPT-4.1 nano61.1%—0.4 s

133

Mixtral 8x7B V0.155.8%—14 s

134

Magistral Medium54.8%—12 s

135

Mistral 7B V0.153.8%—1.7 s

136

Llama 2 13B53.7%—1.7 s

137

Llama 2 7B51.9%—1.0 s

138

GPT-5 nano50.1%—22 s

139

Magistral Small40.0%—6.7 s

140COCommand R33.0%—0.2 s