Gemma 4 31B IT
- Family
- Gemma 4
- Released
- 2026-03-31
- Context
- 256K
- Parameters
- 31B
- Knowledge cutoff
- 2025-01
- Openness
- Open source
- License
- Apache 2.0OSI-approvedPermitted
- Weights
- Available
- Code
- Unknown
Cheapest of 7 routes · Open Inference
Cheapest of 7 routes · Open Inference
About
Instruction-tuned variant of the 31B model with enhanced instruction-following. Maintains #3 Arena leaderboard performance with multimodal capabilities.
Top use-case fit: coding, agents, and build tasks
Coding
Q/$ AProviders(7)
Compare all 7Compare API pricing across 7 tracked provider routes for input and output tokens, batch, and cached reads when available.
| Provider | Input / 1M | Output / 1M | Cache | Route |
|---|---|---|---|---|
| Open Inference | Free | Free | - | Serverless |
| FlexAI | $0.140 | $0.400 | read $0.021 | Serverless |
| Novita AI | $0.140 | $0.400 | - | Serverless |
| OpenRouter | $0.140 | $0.400 | read $0.140 | Serverless |
| GCP Vertex AI | $0.150 | $0.600 | - | Serverless |
| Together AI | $0.390 | $0.970 | - | Serverless |
| Google AI Studio | - | - | - | ServerlessPartial |
Available via routers & gateways(13)
LiteLLM
GatewayOpen-source Python SDK and proxy server that unifies 100+ LLM APIs behind a single OpenAI-compatible interface, with load balancing, cost tracking, and configurable failover.
OpenRouter
HybridUnified hybrid gateway to 500+ models from 80+ providers via a single OpenAI-compatible API, with an optional Auto Router that selects a model per prompt from market spend data.
Portkey
GatewayProduction AI gateway routing to 1,600+ LLMs with failover, load balancing, caching, and guardrails; MIT-licensed open-source core is self-hostable. Now part of Palo Alto Networks as Prisma AIRS AI Gateway.
AIRouter
RouterCommercial LLM router (branded "AI Router" on its site) that picks the best model per request via a Decision API or an OpenAI-compatible Chat Completions endpoint, with a private selection mode that routes on anonymized patterns; sold as BYOK subscription plans priced by routed tokens.
Helicone
GatewayObservability-first AI gateway with routing, caching, rate limiting, and request tracing; Apache-2.0 open-source platform with a managed hosted tier for logging and analytics.
Kong AI Gateway
GatewayMulti-LLM AI gateway built on Kong Gateway 3.x, adding semantic routing, load balancing, guardrails, and MCP traffic analytics as plugins over Kong's existing API management platform.
Capabilities
Benchmark peer barsfor Coding
Benchmark scores(10)
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| 84.3 | diamondObserved 2026-04-18 | — | ||
| 89.2 | AIME 2026 (accuracy)Observed 2026-06-07 | — | ||
| 80.0 | LiveCodeBench v6 (accuracy)Observed 2026-06-07 | — | ||
| 85.2 | From official Gemma 4 model card (accuracy)Observed 2026-06-07 | — | ||
| 1262.0 | vision_style_control overall 2026-10-19 (style control, default view)Observed 2026-10-26 | Configuration: Gemma 4 31B Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena vision style control leaderboard, overall category, published 2026-10-19 (HF lmarena-ai/leaderboard-dataset @9761a3ce, latest split). Board entry 'Gemma 4 31B': rank 45, Arena Score 1261.8 (95% CI 1256.1-1267.6), 40219 votes. Retrieved 2026-10-26. | Source Source: LMArena (lmarena.ai), CC BY 4.0 | |
| 1452.0 | text_style_control overall 2026-10-22 (style control, default view)Observed 2026-10-26 | Configuration: Gemma 4 31B Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena text style control leaderboard, overall category, published 2026-10-22 (HF lmarena-ai/leaderboard-dataset @9761a3ce, latest split). Board entry 'Gemma 4 31B': rank 76, Arena Score 1452.3 (95% CI 1444.8-1459.7), 6137 votes. Retrieved 2026-10-26. | Source Source: LMArena (lmarena.ai), CC BY 4.0 | |
| 1416.0 | text creative_writing 2026-10-22Observed 2026-10-22 | Configuration: Gemma 4 31B Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena Text arena, Creative Writing category (no style control), board published 2026-10-22 (HF lmarena-ai/leaderboard-dataset @a467e6b9, text config, latest split). Board entry 'Gemma 4 31B' (category creative_writing in text/latest-00000-of-00001.parquet): rank 74 of 416, rating 1416.0 (95% CI 1397.3-1434.6), 971 votes. Source: LMArena (CC BY 4.0). Retrieved 2026-10-27. | Source Source: LMArena (lmarena.ai), CC BY 4.0 | |
| 22.0 | WMT26 General MT findings, Figure 3 Avg. Rank (lower is better)Observed 2026-10-27 | Configuration: Gemma 4 - 31B Harness: Contrastive Error Span Annotation (cESA), human annotators Evaluator: WMT26 organizers Confidence: confirmed Notes: WMT26 General MT findings (2026.wmt-1.48), Figure 3: Avg. Rank 22.0 for the system listed as 'Gemma 4 - 31B', averaged over the 23 language pairs it entered; rank 1 is best, so lower is better (not a 0-100 score). Per-pair cESA scores (0-100, higher is better, not comparable across pairs): cs-vi 75.7, cs-de 76.9, en-cs 60.1, en-de 67.8, en-is 43.6, en-zh-Hant 74.3, en-arz 80.5, en-ru 73.7, en-be 49.9, en-lld 3.6, en-se 4.7, en-et 45.8, en-ja 80.5, en-kk 64.4, en-id 75.7, zh-ja 72.2, en-ko 76.2, en-th 80.0, en-lij 6.6, cs-uk 79.3, en-uk 76.3, en-hy 44.0, en-zh-Hans 72.3. WMT26 names 'Gemma 4 - 31B'; mapped to the instruction-tuned slug. Source: WMT26 organizers. Retrieved 2026-10-27. | Source Source: Findings of the WMT26 General Machine Translation Shared Task (Kocmi et al., 2026) | |
| 19.5 | HLE no tools (self-reported, official Gemma 4 model card)Observed 2026-11-05 | Confidence: self-reported Notes: Self-reported by Google on the official Gemma 4 model card (Benchmark Results table, column 'Gemma 4 31B'); the card states results are for instruction-tuned models. Same table on https://huggingface.co/google/gemma-4-31b-it README and https://ai.google.dev/gemma/docs/core/model_card_4. Retrieved 2026-11-05. HLE no tools 19.5%. | ||
| 76.9 | MMMU Pro (self-reported, official Gemma 4 model card)Observed 2026-11-05 | Confidence: self-reported Notes: Self-reported by Google on the official Gemma 4 model card (Benchmark Results table, column 'Gemma 4 31B'); the card states results are for instruction-tuned models. Same table on https://huggingface.co/google/gemma-4-31b-it README and https://ai.google.dev/gemma/docs/core/model_card_4. Retrieved 2026-11-05. MMMU Pro 76.9% (Vision section; no Vision-only setting stated). |
Migration checks
No linked migration route is available for this model yet.