AutoBench Agentic Run 1 - April 2026

The first AutoBench run to measure agentic performance of top LLMs

Past

Date

April 16, 2026

Version

2026-04-16

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)	Adaptive Replanning	Api Workflow	Domain Workflow	Error Handling	Failure Recovery	Multi Step Orchestration	Parallel Execution	Parameter Complexity	Single Tool Call	Tool Selection
Llama 4 Maverick	2.27 (#31)	2.22 (#31)	2.16 (#31)	2.24 (#31)	2.28 (#30)	2.15 (#31)	2.41 (#31)	2.2 (#31)	2.29 (#29)	2.39 (#31)	2.44 (#31)
Gemma 4 26B A4B IT	2.53 (#30)	2.27 (#30)	2.59 (#29)	2.62 (#27)	2.5 (#25)	2.58 (#29)	2.75 (#23)	2.9 (#19)	2.11 (#31)	2.43 (#30)	2.51 (#30)
Deepseek v3.2	2.55 (#29)	2.65 (#23)	2.59 (#28)	2.56 (#29)	2.34 (#29)	2.64 (#26)	2.43 (#30)	2.56 (#29)	2.53 (#11)	2.65 (#26)	2.59 (#26)
Mistral large 2512	2.62 (#28)	2.49 (#27)	2.92 (#12)	2.61 (#28)	2.43 (#28)	2.56 (#30)	2.59 (#29)	2.91 (#18)	2.44 (#22)	2.74 (#23)	2.55 (#29)
Nemotron 3 Nano 30B A3B	2.63 (#27)	2.68 (#22)	2.66 (#26)	2.39 (#30)	2.51 (#24)	2.87 (#22)	2.67 (#26)	2.64 (#27)	2.51 (#15)	2.75 (#21)	2.57 (#28)
Gpt oss 20b	2.65 (#26)	2.48 (#28)	2.91 (#13)	2.65 (#26)	2.47 (#27)	2.76 (#24)	2.67 (#27)	2.53 (#30)	2.46 (#21)	2.82 (#19)	2.81 (#13)
Nova 2 lite v1	2.66 (#25)	2.45 (#29)	2.66 (#27)	2.8 (#21)	2.53 (#22)	2.63 (#27)	2.88 (#16)	2.65 (#25)	2.5 (#18)	2.87 (#14)	2.66 (#20)
Mistral Small 4	2.69 (#24)	2.64 (#24)	2.74 (#23)	2.66 (#25)	2.53 (#23)	2.72 (#25)	2.67 (#28)	2.79 (#24)	2.6 (#8)	2.86 (#16)	2.64 (#25)
Nemotron 3 Super 120B A12B	2.7 (#23)	2.6 (#26)	2.69 (#25)	2.78 (#22)	2.64 (#17)	2.6 (#28)	3 (#10)	2.6 (#28)	2.72 (#4)	2.74 (#22)	2.78 (#14)
Gemma 4 31B IT	2.7 (#22)	2.86 (#14)	2.77 (#19)	2.93 (#16)	2.59 (#19)	2.8 (#23)	2.8 (#21)	2.79 (#22)	2.21 (#30)	2.64 (#28)	2.64 (#23)
MiniMax M2.5	2.72 (#21)	2.68 (#21)	2.57 (#30)	2.71 (#24)	2.49 (#26)	3.17 (#9)	2.88 (#17)	2.92 (#17)	2.41 (#25)	2.64 (#27)	2.59 (#27)
Gpt oss 120b	2.76 (#20)	2.83 (#15)	2.98 (#11)	2.82 (#19)	2.68 (#16)	2.97 (#19)	2.68 (#25)	2.64 (#26)	2.39 (#27)	2.9 (#13)	2.65 (#22)
GPT-5.4 Nano (xhigh)	2.78 (#19)	2.9 (#12)	2.76 (#20)	2.94 (#14)	2.27 (#31)	2.91 (#21)	2.74 (#24)	2.79 (#23)	2.56 (#9)	2.96 (#6)	2.96 (#8)
Gemini 3.1 Flash Lite Preview	2.82 (#18)	2.71 (#20)	2.9 (#15)	2.76 (#23)	2.69 (#15)	3.06 (#15)	2.91 (#15)	3.14 (#10)	2.53 (#10)	2.68 (#25)	2.72 (#17)
Qwen3.5 35B A3B	2.83 (#17)	2.9 (#13)	2.71 (#24)	3 (#12)	2.58 (#20)	2.97 (#18)	2.81 (#20)	3 (#14)	2.35 (#28)	2.9 (#12)	3.08 (#2)
Qwen3.5 122B A10B	2.84 (#16)	2.63 (#25)	2.75 (#21)	2.99 (#13)	2.61 (#18)	3.13 (#10)	2.82 (#19)	3.09 (#11)	2.51 (#14)	2.81 (#20)	2.99 (#6)
Grok 4.1 fast	2.84 (#15)	2.73 (#19)	2.78 (#18)	2.9 (#17)	2.71 (#13)	2.93 (#20)	2.98 (#13)	3.07 (#12)	2.41 (#24)	2.87 (#15)	3.06 (#4)
Gemini 3 Flash Preview	2.89 (#14)	3.07 (#7)	3.06 (#6)	3.11 (#8)	2.7 (#14)	3.02 (#17)	3.05 (#8)	2.95 (#15)	2.72 (#3)	2.64 (#29)	2.68 (#19)
MiniMax M2.7	2.9 (#13)	2.91 (#11)	3.18 (#3)	3.11 (#7)	2.73 (#12)	3.13 (#11)	2.8 (#22)	3.15 (#9)	2.5 (#17)	2.7 (#24)	2.73 (#16)
GPT-5.4 Mini (xhigh)	2.91 (#12)	3.34 (#1)	3.04 (#7)	2.89 (#18)	2.55 (#21)	3.23 (#6)	3.12 (#5)	2.79 (#21)	2.39 (#26)	2.94 (#8)	2.87 (#11)
Claude haiku 4.5	2.92 (#11)	3.06 (#8)	2.84 (#16)	2.94 (#15)	3.07 (#3)	3.22 (#7)	2.99 (#11)	2.88 (#20)	2.47 (#19)	2.82 (#18)	2.65 (#21)
Grok 4.20	2.92 (#10)	2.81 (#17)	2.99 (#10)	2.81 (#20)	2.83 (#9)	3.07 (#14)	3.06 (#7)	2.94 (#16)	2.61 (#7)	3.09 (#2)	2.89 (#10)
GLM 4.7	2.92 (#9)	2.82 (#16)	2.75 (#22)	3.02 (#11)	2.81 (#10)	3.19 (#8)	3.14 (#4)	3.22 (#6)	2.46 (#20)	2.94 (#9)	2.64 (#24)
Kimi K2.5	2.92 (#8)	2.75 (#18)	2.9 (#14)	3.12 (#6)	2.86 (#7)	3.04 (#16)	3.26 (#2)	3.18 (#8)	2.52 (#13)	2.93 (#10)	2.75 (#15)
Mimo V2 Pro	2.99 (#7)	3.17 (#4)	3.11 (#5)	3.06 (#9)	2.81 (#11)	3.25 (#5)	2.98 (#12)	3.26 (#4)	2.43 (#23)	2.86 (#17)	2.84 (#12)
Qwen3.6 Plus	2.99 (#6)	3 (#9)	3.01 (#9)	3.04 (#10)	2.83 (#8)	3.11 (#12)	2.96 (#14)	3.19 (#7)	2.74 (#2)	2.98 (#4)	3.04 (#5)
GPT-5.4 (xhigh)	3.02 (#5)	3.1 (#5)	3.03 (#8)	3.27 (#2)	3.03 (#5)	3.07 (#13)	3.28 (#1)	3.02 (#13)	2.5 (#16)	2.92 (#11)	3.08 (#3)
GLM 5.1	3.06 (#4)	2.97 (#10)	3.13 (#4)	3.17 (#5)	3.07 (#4)	3.3 (#3)	3.04 (#9)	3.26 (#5)	2.62 (#6)	3.04 (#3)	2.72 (#18)
Gemini 3.1 Pro Preview	3.1 (#3)	3.18 (#3)	3.22 (#2)	3.27 (#3)	2.9 (#6)	3.26 (#4)	2.86 (#18)	3.28 (#3)	2.52 (#12)	3.19 (#1)	3.18 (#1)
Claude Sonnet 4.6	3.13 (#2)	3.07 (#6)	2.82 (#17)	3.36 (#1)	3.12 (#1)	3.62 (#1)	3.06 (#6)	3.33 (#2)	2.75 (#1)	2.95 (#7)	2.98 (#7)
Claude Opus 4.6	3.17 (#1)	3.23 (#2)	3.25 (#1)	3.17 (#4)	3.1 (#2)	3.61 (#2)	3.14 (#3)	3.37 (#1)	2.67 (#5)	2.97 (#5)	2.95 (#9)