AutoBench Agronomy LLM Benchmark - December 2025

The first AutoBench run for the Agronomy domain with models Gemini 3 Pro, Gpt 5.1, Grok 4.1, Opus 4.5 and more

Past

Date

December 10, 2025

Version

2025-12-10

Models

New Models

Run data

Overall Comparison Costs Latency P99 Latency Domain

P99 latency breakdown across 10 specialized domains, measured in seconds. This 99th percentile metric reveals worst-case response times, helping you ensure consistent performance for production environments.

Model	Average (All Topics)
Nova lite v1	41.75s (#1)
Nova pro v1	45.94s (#2)
Magistral small 2506	56.42s (#3)
Phi 4	59.91s (#4)
Llama 4 scout	60.13s (#5)
Grok 4.1 fast	64.98s (#6)
Llama 4 Maverick	65.33s (#7)
Claude 3.5 haiku	73.19s (#8)
Gemini 2.5 flash lite	79.85s (#9)
Kimi K2 Instruct	86.47s (#10)
Nemotron nano 9b v2	88.99s (#11)
Gemini 2.5 flash	90.11s (#12)
Grok 3 mini	97.02s (#13)
DeepSeek V3 0324	100.63s (#14)
Gemma 3 27b it	111.58s (#15)
Qwen3 next 80b a3b thinking	126.90s (#16)
Phi 3 mini 128k instruct	142.96s (#17)
Mistral large 2512	143.01s (#18)
Gemini 3 pro preview	143.02s (#19)
GLM 4.5 Air	144.36s (#20)
Gpt oss 120b	152.50s (#21)
Deepseek v3.1	155.68s (#22)
DeepSeek R1 0528	159.18s (#23)
Llama 3.1 nemotron ultra 253b v1	162.15s (#24)
Llama 3.3 nemotron super 49b v1.5	166.08s (#25)
Qwen3 30b a3b instruct 2507	174.47s (#26)
Grok 4.1 fast thinking	176.91s (#27)
Gemini 2.5 pro	186.92s (#28)
GLM 4.5	200.73s (#29)
Grok 4	219.87s (#30)
Gpt 5 mini	224.19s (#31)
Claude opus 4.5	238.10s (#32)
Minimax m2	238.55s (#33)
Qwen3 235B A22B Thinking 2507	254.75s (#34)
Claude sonnet 4.5	283.44s (#35)
Gpt 5	312.34s (#36)
Gpt 5.1	347.66s (#37)
Kimi k2 thinking	360.26s (#38)
Claude haiku 4.5	365.38s (#39)
Deepseek v3.2 exp	381.40s (#40)