AutoBench Run 3 - August 2025

Latest AutoBench run with enhanced metrics including evaluation iterations and fail rates

Latest

Date

August 14, 2025

Version

2025-08-14

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
gpt-5	4.51 (#1)	4.58 (#2)	4.52 (#3)	4.59 (#1)	4.62 (#1)	4.36 (#7)	4.64 (#1)	4.21 (#1)	4.17 (#5)	4.65 (#1)	4.66 (#1)
o3	4.41 (#5)	4.43 (#5)	4.3 (#13)	4.56 (#2)	4.5 (#4)	4.39 (#5)	4.57 (#2)	3.96 (#7)	4.16 (#6)	4.58 (#4)	4.61 (#4)
gpt-oss-120b	4.48 (#3)	4.61 (#1)	4.42 (#6)	4.52 (#3)	4.42 (#7)	4.45 (#2)	4.57 (#3)	4.16 (#3)	4.25 (#2)	4.63 (#2)	4.63 (#2)
gpt-5-mini	4.49 (#2)	4.54 (#3)	4.54 (#1)	4.5 (#4)	4.52 (#3)	4.44 (#4)	4.56 (#4)	4.18 (#2)	4.25 (#1)	4.62 (#3)	4.63 (#3)
Qwen3-235B-A22B-Thinking-2507	4.39 (#6)	4.31 (#8)	4.44 (#5)	4.48 (#5)	4.56 (#2)	4.45 (#3)	4.51 (#5)	3.84 (#10)	3.94 (#8)	4.44 (#9)	4.54 (#5)
gemini-2.5-pro	4.42 (#4)	4.52 (#4)	4.39 (#7)	4.42 (#6)	4.49 (#5)	4.45 (#1)	4.48 (#7)	4.09 (#4)	4.24 (#3)	4.52 (#5)	4.52 (#7)
GLM-4.5	4.18 (#14)	3.89 (#18)	4.26 (#17)	4.38 (#7)	4.42 (#8)	4.32 (#9)	4.47 (#8)	3.48 (#17)	3.47 (#18)	4.48 (#7)	4.49 (#9)
gemini-2.5-flash	4.32 (#8)	4.42 (#6)	4.17 (#21)	4.37 (#8)	4.33 (#14)	4.38 (#6)	4.42 (#11)	4.01 (#6)	4.23 (#4)	4.43 (#10)	4.39 (#15)
o4-mini	4.27 (#10)	4.31 (#10)	4.24 (#18)	4.36 (#9)	4.38 (#12)	4.26 (#12)	4.35 (#15)	3.9 (#8)	3.84 (#10)	4.48 (#6)	4.53 (#6)
gpt-5-nano	4.33 (#7)	4.41 (#7)	4.38 (#8)	4.35 (#10)	4.4 (#10)	4.3 (#11)	4.45 (#9)	3.88 (#9)	4.13 (#7)	4.37 (#15)	4.52 (#8)
deepSeek-R1-0528	4.18 (#12)	3.95 (#17)	4.31 (#12)	4.35 (#11)	4.41 (#9)	4.31 (#10)	4.4 (#12)	3.56 (#13)	3.63 (#12)	4.39 (#14)	4.4 (#13)
grok-4	4.31 (#9)	4.31 (#9)	4.35 (#10)	4.33 (#12)	4.38 (#11)	4.34 (#8)	4.4 (#13)	4.01 (#5)	3.85 (#9)	4.41 (#12)	4.4 (#12)
claude-opus-4-1	4.24 (#11)	4.29 (#11)	4.51 (#4)	4.3 (#13)	4.43 (#6)	4.23 (#14)	4.44 (#10)	3.57 (#12)	3.58 (#13)	4.42 (#11)	4.48 (#10)
claude-sonnet-4	4.17 (#15)	4.19 (#13)	4.36 (#9)	4.3 (#14)	4.33 (#15)	4.25 (#13)	4.35 (#14)	3.55 (#14)	3.48 (#17)	4.4 (#13)	4.39 (#14)
Kimi-K2-Instruct	4.18 (#13)	4.12 (#14)	4.54 (#2)	4.29 (#15)	4.35 (#13)	4.19 (#16)	4.5 (#6)	3.4 (#21)	3.52 (#15)	4.46 (#8)	4.41 (#11)
gpt-4.1	4.17 (#16)	4.24 (#12)	4.32 (#11)	4.26 (#16)	4.19 (#19)	4.21 (#15)	4.23 (#18)	3.74 (#11)	3.79 (#11)	4.27 (#16)	4.32 (#16)
GLM-4.5-Air	3.98 (#20)	3.8 (#22)	3.99 (#25)	4.19 (#17)	4.19 (#18)	4.03 (#21)	4.29 (#16)	3.42 (#20)	3.34 (#23)	4.25 (#18)	4.27 (#17)
llama-3_1-Nemotron-Ultra-253B-v1	4.02 (#18)	3.77 (#24)	4.27 (#16)	4.18 (#18)	4.22 (#16)	4.09 (#18)	4.26 (#17)	3.47 (#18)	3.43 (#20)	4.23 (#21)	4.24 (#19)
grok-3-mini	4.06 (#17)	4.02 (#16)	4.18 (#20)	4.16 (#19)	4.21 (#17)	4.16 (#17)	4.21 (#19)	3.51 (#15)	3.49 (#16)	4.26 (#17)	4.26 (#18)
Qwen3-14B	3.98 (#21)	3.83 (#21)	4.28 (#15)	4.11 (#20)	4.09 (#24)	4.02 (#22)	4.15 (#23)	3.44 (#19)	3.34 (#22)	4.23 (#20)	4.16 (#22)
gemma-3-27b-it	3.88 (#25)	3.57 (#27)	4.29 (#14)	4.11 (#21)	4.17 (#20)	3.97 (#24)	4.18 (#20)	3.04 (#30)	3.08 (#29)	4.19 (#22)	4.2 (#20)
Qwen3-30B-A3B	3.95 (#22)	3.8 (#23)	4.17 (#22)	4.09 (#22)	4.16 (#22)	3.94 (#25)	4.14 (#24)	3.49 (#16)	3.39 (#21)	4.13 (#24)	4.15 (#23)
gemini-2.5-flash-lite	4.02 (#19)	4.11 (#15)	4.15 (#23)	4.08 (#23)	4.17 (#21)	4.05 (#19)	4.16 (#22)	3.34 (#23)	3.55 (#14)	4.24 (#19)	4.19 (#21)
deepSeek-V3-0324	3.95 (#23)	3.88 (#19)	4.19 (#19)	4.07 (#24)	4.06 (#25)	4.05 (#20)	4.09 (#25)	3.37 (#22)	3.44 (#19)	4.14 (#23)	4.1 (#25)
Llama-3_3-Nemotron-Super-49B-v1	3.88 (#24)	3.83 (#20)	4.05 (#24)	4.04 (#25)	4.12 (#23)	3.99 (#23)	4.16 (#21)	3.04 (#29)	3.27 (#25)	4.12 (#25)	4.13 (#24)
magistral-small-2506	3.71 (#27)	3.74 (#25)	3.23 (#33)	3.92 (#26)	3.89 (#28)	3.84 (#27)	3.97 (#26)	3.22 (#24)	3.28 (#24)	4 (#26)	3.94 (#27)
mistral-large-2411	3.71 (#26)	3.5 (#28)	3.97 (#26)	3.83 (#27)	3.93 (#27)	3.8 (#29)	3.92 (#27)	3.11 (#27)	3.11 (#27)	3.99 (#27)	3.95 (#26)
claude-3.5-haiku	3.59 (#31)	3.47 (#30)	3.86 (#29)	3.74 (#28)	4 (#26)	3.74 (#30)	3.87 (#28)	2.82 (#33)	2.78 (#33)	3.74 (#33)	3.81 (#30)
phi-4	3.66 (#28)	3.48 (#29)	3.97 (#27)	3.73 (#29)	3.85 (#29)	3.66 (#31)	3.82 (#31)	3.13 (#26)	3.2 (#26)	3.87 (#28)	3.85 (#28)
nova-lite-v1	3.54 (#32)	3.32 (#33)	3.78 (#31)	3.71 (#30)	3.77 (#32)	3.51 (#33)	3.76 (#32)	2.99 (#31)	2.95 (#31)	3.82 (#30)	3.75 (#32)
llama-4-maverick	3.64 (#29)	3.59 (#26)	3.74 (#32)	3.7 (#31)	3.78 (#31)	3.83 (#28)	3.82 (#30)	3.13 (#25)	3.1 (#28)	3.83 (#29)	3.79 (#31)
llama-4-Scout-17B-16E-Instruct	3.61 (#30)	3.37 (#31)	3.86 (#28)	3.66 (#32)	3.83 (#30)	3.85 (#26)	3.84 (#29)	3.05 (#28)	3 (#30)	3.82 (#31)	3.84 (#29)
nova-pro-v1	3.49 (#33)	3.36 (#32)	3.84 (#30)	3.55 (#33)	3.72 (#33)	3.53 (#32)	3.63 (#33)	2.96 (#32)	2.85 (#32)	3.75 (#32)	3.61 (#33)