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

Track the cost efficiency of LLM models across 10 specialized domains. Costs are measured in cents per response, helping you identify the most economical models for your specific use case.

Model	Average (All Topics)	Coding	Creative Writing	Current News	General Culture	Grammar	History	Logics	Math	Science	Technology
nova-lite-v1	0.02 (#1)	0.03 (#1)	0.01 (#1)	0.02 (#1)	0.01 (#1)	0.01 (#1)	0.02 (#1)	0.02 (#1)	0.02 (#1)	0.02 (#1)	0.02 (#1)
gemma-3-27b-it	0.03 (#3)	0.04 (#3)	0.02 (#2)	0.03 (#3)	0.02 (#3)	0.02 (#3)	0.03 (#3)	0.03 (#3)	0.04 (#3)	0.03 (#3)	0.03 (#3)
phi-4	0.02 (#2)	0.03 (#2)	0.02 (#3)	0.02 (#2)	0.02 (#2)	0.02 (#2)	0.02 (#2)	0.02 (#2)	0.04 (#2)	0.02 (#2)	0.02 (#2)
gemini-2.5-flash-lite	0.11 (#10)	0.16 (#10)	0.02 (#4)	0.04 (#6)	0.03 (#4)	0.04 (#4)	0.07 (#10)	0.28 (#14)	0.30 (#14)	0.05 (#8)	0.08 (#10)
llama-4-Scout-17B-16E-Instruct	0.04 (#4)	0.05 (#4)	0.03 (#5)	0.04 (#5)	0.04 (#7)	0.04 (#5)	0.04 (#5)	0.04 (#4)	0.05 (#4)	0.04 (#4)	0.04 (#4)
Llama-3_3-Nemotron-Super-49B-v1	0.05 (#5)	0.06 (#5)	0.03 (#6)	0.04 (#4)	0.04 (#6)	0.04 (#6)	0.04 (#4)	0.07 (#6)	0.06 (#5)	0.04 (#5)	0.04 (#5)
llama-4-maverick	0.05 (#6)	0.07 (#6)	0.03 (#7)	0.04 (#7)	0.03 (#5)	0.04 (#7)	0.05 (#6)	0.07 (#5)	0.07 (#6)	0.04 (#6)	0.04 (#6)
Qwen3-14B	0.08 (#8)	0.10 (#7)	0.04 (#8)	0.05 (#8)	0.04 (#9)	0.06 (#9)	0.05 (#8)	0.19 (#12)	0.19 (#10)	0.05 (#9)	0.05 (#8)
Qwen3-30B-A3B	0.08 (#7)	0.12 (#8)	0.04 (#9)	0.05 (#9)	0.04 (#8)	0.05 (#8)	0.05 (#7)	0.15 (#8)	0.19 (#9)	0.05 (#7)	0.05 (#7)
grok-3-mini	0.09 (#9)	0.14 (#9)	0.05 (#10)	0.07 (#10)	0.05 (#10)	0.07 (#10)	0.07 (#9)	0.15 (#7)	0.13 (#7)	0.07 (#10)	0.07 (#9)
gpt-oss-120b	0.14 (#12)	0.20 (#12)	0.08 (#11)	0.14 (#13)	0.10 (#13)	0.09 (#11)	0.12 (#13)	0.16 (#9)	0.21 (#11)	0.12 (#13)	0.14 (#13)
deepSeek-V3-0324	0.12 (#11)	0.18 (#11)	0.08 (#12)	0.11 (#12)	0.08 (#12)	0.10 (#12)	0.10 (#12)	0.18 (#11)	0.16 (#8)	0.10 (#12)	0.10 (#12)
magistral-small-2506	0.20 (#14)	0.21 (#13)	0.10 (#13)	0.10 (#11)	0.07 (#11)	0.11 (#13)	0.10 (#11)	0.61 (#18)	0.52 (#16)	0.10 (#11)	0.10 (#11)
GLM-4.5-Air	0.36 (#18)	0.65 (#19)	0.12 (#14)	0.20 (#17)	0.14 (#15)	0.31 (#18)	0.19 (#15)	0.67 (#19)	0.81 (#20)	0.24 (#18)	0.27 (#18)
nova-pro-v1	0.18 (#13)	0.29 (#14)	0.15 (#15)	0.17 (#15)	0.13 (#14)	0.18 (#15)	0.16 (#14)	0.17 (#10)	0.25 (#13)	0.15 (#14)	0.15 (#14)
gemini-2.5-flash	0.45 (#20)	1.01 (#22)	0.15 (#16)	0.39 (#21)	0.24 (#19)	0.33 (#20)	0.37 (#20)	0.40 (#16)	0.65 (#17)	0.42 (#22)	0.43 (#22)
Kimi-K2-Instruct	0.24 (#15)	0.29 (#15)	0.17 (#17)	0.24 (#18)	0.22 (#18)	0.17 (#14)	0.29 (#18)	0.28 (#13)	0.22 (#12)	0.24 (#17)	0.23 (#17)
gpt-5-nano	0.24 (#16)	0.32 (#16)	0.19 (#18)	0.18 (#16)	0.16 (#17)	0.26 (#17)	0.22 (#17)	0.35 (#15)	0.38 (#15)	0.17 (#15)	0.19 (#16)
llama-3_1-Nemotron-Ultra-253B-v1	0.35 (#17)	0.52 (#18)	0.21 (#19)	0.17 (#14)	0.14 (#16)	0.23 (#16)	0.21 (#16)	0.77 (#22)	0.84 (#21)	0.21 (#16)	0.19 (#15)
GLM-4.5	0.63 (#22)	1.22 (#24)	0.22 (#20)	0.37 (#20)	0.30 (#21)	0.37 (#21)	0.38 (#21)	1.21 (#24)	1.44 (#25)	0.38 (#21)	0.41 (#21)
deepSeek-R1-0528	0.64 (#24)	1.31 (#26)	0.24 (#21)	0.36 (#19)	0.29 (#20)	0.32 (#19)	0.35 (#19)	1.36 (#25)	1.57 (#26)	0.34 (#19)	0.35 (#19)
Qwen3-235B-A22B-Thinking-2507	0.42 (#19)	0.45 (#17)	0.28 (#22)	0.44 (#22)	0.36 (#22)	0.41 (#22)	0.38 (#22)	0.67 (#21)	0.66 (#18)	0.37 (#20)	0.37 (#20)
gpt-5-mini	0.63 (#23)	0.84 (#20)	0.36 (#23)	0.56 (#23)	0.42 (#23)	0.50 (#24)	0.52 (#23)	0.89 (#23)	1.10 (#23)	0.56 (#23)	0.58 (#23)
mistral-large-2411	0.61 (#21)	0.96 (#21)	0.46 (#24)	0.61 (#24)	0.46 (#24)	0.46 (#23)	0.61 (#24)	0.57 (#17)	0.73 (#19)	0.58 (#24)	0.59 (#24)
gpt-4.1	0.91 (#27)	1.28 (#25)	0.48 (#25)	0.80 (#27)	0.53 (#25)	0.72 (#26)	0.75 (#26)	1.40 (#26)	1.69 (#27)	0.76 (#26)	0.75 (#26)
claude-3.5-haiku	0.83 (#25)	1.43 (#27)	0.60 (#26)	0.78 (#26)	0.69 (#27)	0.70 (#25)	0.83 (#27)	0.67 (#20)	0.87 (#22)	0.76 (#27)	0.77 (#27)
o4-mini	0.87 (#26)	1.03 (#23)	0.63 (#27)	0.71 (#25)	0.61 (#26)	0.73 (#27)	0.69 (#25)	1.60 (#29)	1.25 (#24)	0.67 (#25)	0.74 (#25)
gemini-2.5-pro	1.59 (#28)	2.77 (#29)	0.73 (#28)	1.51 (#29)	1.12 (#29)	1.22 (#30)	1.49 (#29)	1.52 (#27)	2.21 (#29)	1.61 (#30)	1.52 (#29)
o3	1.85 (#30)	1.83 (#28)	0.94 (#29)	1.52 (#30)	1.01 (#28)	1.05 (#28)	1.34 (#28)	3.76 (#30)	5.16 (#30)	1.26 (#28)	1.36 (#28)
claude-sonnet-4	1.71 (#29)	3.74 (#30)	0.99 (#30)	1.47 (#28)	1.16 (#30)	1.13 (#29)	1.59 (#30)	1.52 (#28)	1.81 (#28)	1.52 (#29)	1.55 (#30)
grok-4	2.92 (#31)	5.10 (#31)	1.49 (#31)	2.30 (#31)	1.62 (#31)	2.76 (#31)	2.20 (#31)	5.12 (#31)	5.97 (#31)	2.11 (#31)	2.26 (#31)
gpt-5	4.37 (#32)	6.01 (#32)	2.75 (#32)	3.79 (#32)	3.07 (#32)	3.33 (#32)	3.68 (#32)	6.20 (#32)	7.59 (#32)	3.76 (#32)	3.82 (#32)
claude-opus-4-1	9.13 (#33)	18.54 (#33)	5.81 (#33)	7.95 (#33)	7.09 (#33)	6.34 (#33)	9.15 (#33)	7.76 (#33)	8.97 (#33)	8.11 (#33)	8.62 (#33)