Throughput (tokens / sec)
avg. 58.6tokens/s
Multimodal long-context work with thinking control
Run instantly. Pay only for what you use.
Gemma 4 31B is the dense flagship of Google DeepMind's Gemma 4 family, released under Apache 2.0: 30.7B parameters, a 256K-token context window, native text + image input, and pre-training across 140+ languages, with a configurable thinking mode.
What sets it apart is human preference: at 1452 ELO on LMArena it is the top-ranked open dense model — ahead of open models several times its size — so answers read well to users, not just to benchmarks.
With thinking enabled it posts 85.2% on MMLU-Pro, 89.2 on AIME 2026, and 80.0 on LiveCodeBench v6, with strict function calling and JSON output for production pipelines.
1452LMArena ELO
The top-ranked open dense model in human preference evaluations.
256Ktoken context window
Holds long documents, retrieved passages, and multi-turn sessions in one prompt.
140+languages in pre-training
35+ supported out of the box — one deployment serves a global user base.
Gemma 4 31B is strongest where answer quality is judged by people — assistants, document analysis, and multilingual products that need flagship reasoning with native vision.
85.2% on MMLU-Pro and 89.2 on AIME 2026 with thinking enabled — reasoning depth on demand, direct replies with it off.
76.9 on MMMU-Pro — screenshots, PDFs, and diagrams are read alongside text, no separate vision model to route to.
80.0 on LiveCodeBench v6 — reviews, patches, and test generation with 256K of repository context in scope.
Strict function calling and JSON mode — tool arguments and extraction pipelines that parse the first time.
Human-facing assistants and chat products
The top Arena-ranked open dense model — responses users actually prefer, in 35+ languages out of the box.
Multimodal document analysis at long context
Native vision at 76.9 on MMMU-Pro over a 256K window — contracts, reports, and image-heavy files in one prompt.
Math-heavy and structured reasoning
89.2 on AIME 2026 with thinking enabled — step-by-step depth on demand, direct replies with it off.
Document and image extraction to JSON
Strict function calling and JSON mode return schema-valid output for ticket, form, and invoice pipelines.
Throughput-first, cost-dominated fleets
Every token runs all 30.7B parameters. When per-token cost dominates, Gemma 4 26B A4B keeps near-flagship quality at 3.8B-active pricing.
The heaviest legacy-code refactors
Community testing shows it trailing dedicated coding agents on complex architectural work — compare DeepSeek V4 Flash or Qwen 3.6 27B for repo-scale refactors.
Audio input or speech understanding
Inputs are text and image only. Transcribe audio upstream before sending it to the model.
Compare Entrim’s pricing, powered by an optimized inference runtime, against other providers using the same token counts per request.
avg. 58.6tokens/s
avg. 850ms
85.2%MMLU-Pro
Graduate-level knowledge and reasoning across 14 domains.
89.2AIME 2026
Competition mathematics measuring multi-step reasoning.
80.0LiveCodeBench v6
Competitive coding problems unseen at training time.
Code examples
curl -X POST "https://api.entrim.ai/v1/chat/completions" \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $ENTRIM_API_KEY" \ -d '{ "model": "google/gemma-4-31B-it", "messages": [ { "role": "user", "content": "Why do they call it a building if it is already built?" } ] }'SDKs & docs
Use familiar SDK patterns with quickstart examples for common production setups.
OpenAI-compatible
Keep the request format your team already knows.
No cold starts
Production requests are served without model spin-up delays.
Gemma 4 quality at 4B-active speed
Dense agentic coding with thinking control
High-throughput agentic work with 1M context