AutoBench Run 2 - April 2025

Second major AutoBench run with o4-mini, GPT-4.1-mini, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet:thinking, etc.

Past

Date

April 25, 2025

Version

2025-04-25

Models

New Models

Run data

Overall Comparison Costs Latency P99 Latency Domain

Evaluate model performance across 10 specialized knowledge domains with normalized scores. Identify which models excel in specific areas—whether you need strong coding capabilities, creative writing prowess, or expertise in science and mathematics. Average scores reflect overall domain competency.

Model	Average (All Topics)	Coding	Creative Writing	Current News	General Culture	Grammar	History	Logics	Math	Science	Technology
Nova pro v1	3.83 (#25)	3.81 (#23)	4.06 (#18)	3.78 (#24)	3.9 (#24)	3.91 (#23)	3.82 (#24)	3.74 (#23)	3.56 (#21)	3.86 (#24)	3.86 (#24)
Nova lite v1	3.89 (#23)	3.73 (#24)	3.86 (#23)	3.86 (#23)	4.04 (#23)	3.9 (#24)	4.02 (#23)	3.77 (#22)	3.56 (#20)	4.02 (#23)	4.05 (#23)
Claude 3.5 Haiku (20241022)	3.99 (#22)	4 (#15)	4.2 (#14)	4.04 (#18)	4.11 (#17)	3.98 (#20)	4.15 (#15)	3.85 (#20)	3.44 (#24)	4.05 (#22)	4.07 (#21)
GPT-4o Mini	4 (#20)	3.97 (#19)	4.2 (#15)	4 (#21)	4.1 (#18)	3.97 (#21)	4.03 (#21)	3.82 (#21)	3.79 (#15)	4.07 (#21)	4.07 (#22)
Mistral Small 24B Instruct	3.88 (#24)	3.86 (#21)	3.42 (#24)	4.01 (#20)	4.05 (#22)	3.94 (#22)	4.02 (#22)	3.66 (#24)	3.59 (#19)	4.09 (#20)	4.08 (#20)
Llama 4 Maverick 17B Instruct	4 (#19)	3.98 (#17)	4.04 (#19)	3.99 (#22)	4.05 (#21)	4.1 (#17)	4.1 (#18)	3.86 (#19)	3.64 (#18)	4.1 (#19)	4.1 (#18)
DeepSeek V3 0324	4.16 (#13)	4.25 (#7)	4.33 (#10)	4.17 (#10)	4.17 (#12)	4.22 (#11)	4.18 (#13)	4.07 (#6)	3.97 (#7)	4.11 (#18)	4.13 (#16)
DeepSeek V3	4.09 (#16)	4.01 (#14)	4.32 (#11)	4.08 (#13)	4.14 (#16)	4.11 (#15)	4.06 (#20)	4.04 (#9)	3.91 (#9)	4.13 (#17)	4.12 (#17)
Llama 4 Scout 17B 16E Instruct	4 (#21)	3.88 (#20)	4.04 (#20)	4.04 (#19)	4.1 (#19)	4.09 (#18)	4.11 (#17)	3.89 (#17)	3.53 (#22)	4.14 (#16)	4.09 (#19)
Llama 3.3 70B Instruct	4.02 (#18)	3.83 (#22)	4.02 (#21)	4.07 (#15)	4.17 (#14)	4.1 (#16)	4.13 (#16)	3.93 (#14)	3.52 (#23)	4.15 (#15)	4.21 (#13)
Mistral Large 2411	4.05 (#17)	3.98 (#18)	4.19 (#16)	4.07 (#16)	4.08 (#20)	4.05 (#19)	4.09 (#19)	3.87 (#18)	3.88 (#11)	4.17 (#14)	4.18 (#14)
Grok-2	4.1 (#15)	4.12 (#11)	4.17 (#17)	4.08 (#14)	4.17 (#13)	4.16 (#13)	4.16 (#14)	3.92 (#15)	3.87 (#13)	4.19 (#13)	4.14 (#15)
o3-mini	4.26 (#6)	4.44 (#3)	4.35 (#9)	4.09 (#12)	4.2 (#11)	4.23 (#10)	4.21 (#11)	4.32 (#2)	4.41 (#2)	4.21 (#12)	4.25 (#11)
Qwen Plus	4.17 (#12)	4.23 (#8)	4.3 (#13)	4.06 (#17)	4.17 (#15)	4.19 (#12)	4.21 (#12)	4.1 (#5)	4.03 (#6)	4.22 (#11)	4.24 (#12)
Gemini 2.0 Flash	4.16 (#14)	4.18 (#9)	3.99 (#22)	4.18 (#9)	4.28 (#10)	4.24 (#9)	4.3 (#10)	3.97 (#12)	3.85 (#14)	4.25 (#10)	4.29 (#9)
Claude 3.7 Sonnet	4.2 (#9)	4.27 (#6)	4.41 (#5)	4.15 (#11)	4.31 (#8)	4.14 (#14)	4.34 (#6)	3.96 (#13)	3.87 (#12)	4.29 (#9)	4.3 (#8)
Claude 3.7 Sonnet	4.39 (#3)	4.27 (#6)	4.41 (#5)	4.15 (#11)	4.31 (#8)	4.14 (#14)	4.34 (#6)	3.96 (#13)	3.87 (#12)	4.29 (#9)	4.3 (#8)
Llama 3.1 Nemotron 70B Instruct	4.18 (#11)	4.1 (#12)	4.3 (#12)	4.2 (#8)	4.32 (#7)	4.27 (#8)	4.32 (#9)	3.99 (#10)	3.68 (#17)	4.3 (#8)	4.29 (#10)
GPT-4.1 Mini	4.34 (#4)	4.42 (#4)	4.41 (#6)	4.22 (#7)	4.3 (#9)	4.44 (#4)	4.32 (#8)	4.3 (#3)	4.34 (#3)	4.3 (#7)	4.4 (#3)
Llama 3.1 Nemotron Ultra 253B v1	4.26 (#7)	4.17 (#10)	4.38 (#7)	4.33 (#4)	4.33 (#6)	4.36 (#5)	4.34 (#7)	4.06 (#7)	3.91 (#10)	4.31 (#6)	4.36 (#6)
Gemma 3 27b it	4.2 (#10)	3.98 (#16)	4.35 (#8)	4.29 (#6)	4.36 (#4)	4.33 (#6)	4.38 (#5)	3.9 (#16)	3.7 (#16)	4.33 (#5)	4.34 (#7)
DeepSeek R1	4.26 (#8)	4.05 (#13)	4.46 (#3)	4.29 (#5)	4.35 (#5)	4.32 (#7)	4.39 (#4)	3.97 (#11)	3.95 (#8)	4.35 (#4)	4.39 (#4)
Grok-3 Beta	4.34 (#5)	4.33 (#5)	4.47 (#2)	4.36 (#3)	4.43 (#3)	4.54 (#2)	4.45 (#3)	4.05 (#8)	4.07 (#5)	4.42 (#3)	4.36 (#5)
Gemini 2.5 Pro Preview	4.46 (#2)	4.5 (#2)	4.42 (#4)	4.48 (#2)	4.59 (#1)	4.53 (#3)	4.6 (#2)	4.17 (#4)	4.17 (#4)	4.56 (#2)	4.59 (#2)
o4-mini	4.57 (#1)	4.55 (#1)	4.51 (#1)	4.57 (#1)	4.59 (#2)	4.6 (#1)	4.61 (#1)	4.48 (#1)	4.57 (#1)	4.67 (#1)	4.61 (#1)