Atlas

Models

← All models

Gemma 4 31B IT

Google DeepMind · 2026-04-02 · 32.7B parameters

Gemma 4 31B IT is Google DeepMind's instruction-tuned dense Gemma 4 checkpoint, released on April 2, 2026 under Apache 2.0; its official checkpoint index reports 32,682,372,656 total parameters. It accepts text and image input, can analyze video as sampled frame sequences, generates text across a 262,144-token context window, and exposes a caller-controlled thinking mode. Google states the pretraining-data cutoff only as January 2025, so the structured knowledge_cutoff remains null.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo373
AA-LCR62.0
AA-Omniscience Index-45.4
ALE-Bench1394
Artificial Analysis Agentic Index14.4
Artificial Analysis Coding Index43.4
Artificial Analysis Intelligence Index29.4
Artificial Analysis Omniscience Accuracy19.9
Artificial Analysis Omniscience Hallucination Rate81.6
Artificial Analysis Openness Index38.9
AutomationBench (Overall)1.7
BRIDGE Medical (chain-of-thought)45.9
BRIDGE Medical (few-shot)54.9
BRIDGE Medical (zero-shot)46.7
BullshitBench v2 Leaderboard25.0
Buyout Game Bradley-Terry Rating1522
CadQueryEval24.0
CadQueryEval — Bounding Box28.0
CadQueryEval — Chamfer Distance36.0
CadQueryEval — Code Execution44.0
CadQueryEval — Component Count40.0
CadQueryEval — Hausdorff 95p28.0
CadQueryEval — STL Creation44.0
CadQueryEval — Volume28.0
CadQueryEval — Watertight Mesh36.0
Capability148.5
CaseLaw52.6
Creative Writing v3 Elo1409
Creative Writing v3 Rubric Score16.0
CRED v0.9236.5
Loading Atlas data…