Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

G9v3 3B

Other · 3B

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Agents52.2—/267✱
Tool use55.6—/81✱
Reasoning47—/178✱
Mathematics47.2—/140✱
Science38.9—/122✱
Expert reasoning41—/0✱
Coding45.6—/164✱
Agentic coding44.5—/157✱
Code generation38.1—/77✱

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for G9v3 3B, left for the other.

GPT-4owon 6 · lost 0
Mistral Medium 3.5won 6 · lost 0
GPT-4.1 Nanowon 7 · lost 0
GPT-4o Miniwon 7 · lost 0
Qwen3.5 0.8Bwon 6 · lost 1
Grok 4.3won 6 · lost 1
Mistral Large 3won 5 · lost 1
Qwen3.6 27Bwon 5 · lost 1
Grok Build 0.1won 5 · lost 1
Gemini 3.5 Flash Litewon 5 · lost 2
Command Awon 5 · lost 2
Gemma 4 E4Bwon 4 · lost 2
Claude 3.5 Haikuwon 4 · lost 2
DeepSeek V3won 4 · lost 3
GPT-5.4 Nanowon 4 · lost 3
GPT-5.4 Miniwon 4 · lost 3
Inklingwon 5 · lost 4
MiniMax M3won 5 · lost 4
Claude 3.5 Sonnetwon 3 · lost 3
Gemini 3.1 Flash Litewon 3 · lost 3
GPT OSS 120Bwon 3 · lost 3
Inkling Smallwon 3 · lost 3
MiniMax M2.7won 3 · lost 3
Mimo V2.5won 3 · lost 3
Nemotron 3 Ultrawon 4 · lost 4
Granite 4.2 30Bwon 4 · lost 4
K2 Horizon 0.9Bwon 3 · lost 4
OpenBMB 1Bwon 3 · lost 4
GPT-4.1 Miniwon 3 · lost 4
GPT-5 Miniwon 3 · lost 4
Qwen3.7 Maxwon 3 · lost 4
Claude Opus 4won 2 · lost 4
Claude 3.7 Sonnetwon 2 · lost 4
Kimi K2.7 Codewon 2 · lost 4
Kimi K2.5won 2 · lost 4
Qwen3.5 4Bwon 3 · lost 6
O3 Miniwon 2 · lost 5
Qwen3.5 2Bwon 2 · lost 5
Gemini 3.5 Flashwon 2 · lost 5
Qwen3.6 Pluswon 2 · lost 5
Claude Haiku 4.5won 2 · lost 5
GPT-4.1won 2 · lost 5
GLM-4.6won 2 · lost 5
Grok 4.1 Fastwon 2 · lost 5
GLM-5.3 Flashwon 2 · lost 5
Granite 4.2 8Bwon 2 · lost 6
Gemma 4 31Bwon 2 · lost 6
Qwen3.5 9Bwon 2 · lost 6
Nemotron 3 Superwon 2 · lost 6
Qwen3.6 35B A3Bwon 2 · lost 6
G9v3 39A5Bwon 2 · lost 6
K2 Horizon 32Bwon 2 · lost 6
Muse Glimmer-30Bwon 2 · lost 6
Gemini 2.5 Flashwon 2 · lost 6
Granite 4.2 3Bwon 2 · lost 7
K2 Horizon 3.7Bwon 2 · lost 7
DeepSeek R1won 1 · lost 5
Muse Sparkwon 1 · lost 5
GPT-5.1won 1 · lost 5
GPT-5won 1 · lost 5
Mimo V2.5 Prowon 1 · lost 5
Qwen3.8 Flash Nextwon 1 · lost 5
Gemini 2.5 Prowon 1 · lost 6
Gemma 4 12Bwon 1 · lost 6
Gemini 3.1 Prowon 1 · lost 6
Claude Sonnet 4won 1 · lost 6
Grok 4.5won 1 · lost 6
Gemini 3.6 Flashwon 1 · lost 6
Kimi K2.6won 1 · lost 6
Kimi K2won 1 · lost 6
Muse Spark 1.1won 1 · lost 6
K2 Horizon 7Bwon 1 · lost 7
DeepSeek V4 Flashwon 1 · lost 7
Claude Opus 4.5won 1 · lost 7
Qwen3.8 27Bwon 1 · lost 8
Claude Opus 4.6won 0 · lost 6
Grok 3 Miniwon 0 · lost 6
GLM-4.7won 0 · lost 6
GLM-5.1won 0 · lost 6
GPT-6 Lunawon 0 · lost 6
Muse Spark 1.2won 0 · lost 6
GPT-6 Solwon 0 · lost 6
GPT-6 Astrawon 0 · lost 6
Muse Spark 1.3won 0 · lost 6
Grok 4.7won 0 · lost 6
Claude Opus 5.5won 0 · lost 6
DeepSeek V3.2won 0 · lost 7
GPT-5.4won 0 · lost 7
Gemini 3.7 Flashwon 0 · lost 7
GPT-5.5won 0 · lost 7
Claude Sonnet 4.5won 0 · lost 7
Claude Sonnet 4.6won 0 · lost 7
Claude Opus 4.7won 0 · lost 7
Gemini 3.8 Flashwon 0 · lost 7
Grok 4won 0 · lost 7
Claude Opus 4.8won 0 · lost 7
Gemini 3 Prowon 0 · lost 7
Grok 4.6won 0 · lost 7
Qwen3.8 Maxwon 0 · lost 7
Claude Fable 5.1won 0 · lost 7
Claude Fable 5won 0 · lost 7
O3won 0 · lost 8
O4 Miniwon 0 · lost 8
GPT-5.2won 0 · lost 8
K2 Horizon 375B A23Bwon 0 · lost 8
DeepSeek V4 Prowon 0 · lost 8
DeepSeek V4.1 Flashwon 0 · lost 8
GPT-5.6 Solwon 0 · lost 8
Kimi K3won 0 · lost 8
GLM-5.3won 0 · lost 8
Claude Opus 5won 0 · lost 8
GLM-5.2won 0 · lost 9
GPT-5.6 Terrawon 0 · lost 9
GPT-5.6 Lunawon 0 · lost 9
Claude Sonnet 5won 0 · lost 9

§ 3 · Sources

Where the numbers come from

1 publication, 7 figures. Every one links to the page it was read from.

Terminal-Bench 2.1 6%SciCode 17.7%GPQA Diamond 43.8%SWE-bench Verified 16.4%HMMT Feb 2026 34.1%HLE 4.5%BFCL v4 47.9%

Badge

PublicAI Index badge for G9v3 3B[![PublicAI Index](https://publicai.io/model-index/badge?model=g9v3-3b)](https://publicai.io/model-index/m/g9v3-3b)

✱ Placed by a one-off publication, not a board that re-ran the model. ✱✱ marks a figure the model’s own publisher printed, which counts for less again. Scores are 0–100 on the PublicAI Index scale, 50 = the average of the models each source lists. Scores belong to their publishers.

Snapshot 2026-09-27 · the full index · JSON API