Gemini 3.1 Pro

2026-02-19 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 3.1 Pro in preview, a reasoning model that reaches 77.1% on ARC-AGI-2, more than double Gemini 3 Pro, available via the API, Vertex AI and the Gemini app.

What changes

  • Clear progress in core reasoning: ARC-AGI-2 verified at 77.1%, more than double 3 Pro.
  • Better results in agentic tasks (MCP Atlas, BrowseComp) and competitive programming (LiveCodeBench Pro).
  • Generation of animated SVGs in code, synthesis of complex systems, interactive design.
  • Launched in preview to validate the changes before general availability.
  • Variant gemini-3.1-pro-preview-customtools published in the API changelog.

### Specifications - Context: up to 1 million tokens of input; output up to 64,000 tokens (model card). - Modalities: text, images, audio, video and code repositories as input, text as output. - Availability: Gemini API (AI Studio, Gemini CLI, Antigravity, Android Studio), Vertex AI, Gemini Enterprise, Gemini app, NotebookLM. - API identifier: gemini-3.1-pro-preview. Cutoff date: not extracted from the model card. ### Pricing | Tier | Input / 1M tokens | Output / 1M tokens | |---|---|---| | Standard (≤ 200k tokens) | USD 2.00 | USD 12.00 | | Standard (> 200k tokens) | USD 4.00 | USD 18.00 | | Batch / Flex (≤ 200k) | USD 1.00 | USD 6.00 | Cache: USD 0.20 (≤ 200k) and USD 0.40 (> 200k), plus storage at USD 4.50 per million tokens per hour. Pricing recorded from the current pricing page. ### Benchmarks | Benchmark (scores published by the vendor) | Gemini 3.1 Pro | Gemini 3 Pro | Opus 4.6 | GPT-5.2 | |---|---|---|---|---| | Humanity's Last Exam (no tools) | 44.4% | 37.5% | 40.0% | 34.5% | | ARC-AGI-2 | 77.1% | 31.1% | 68.8% | 52.9% | | GPQA Diamond | 94.3% | 91.9% | 91.3% | 92.4% | | SWE-Bench Verified | 80.6% | 76.2% | 80.8% | 80.0% | | Terminal-Bench 2.0 (Terminus-2) | 68.5% | 56.9% | 65.4% | 54.0% | | BrowseComp | 85.9% | 59.2% | 84.0% | 65.8% | | MCP Atlas | 69.2% | 54.1% | 59.5% | 60.6% | ### Key takeaways The best-performing Pro model of generation 3 in reasoning at the time of its launch, but still in preview: to be validated before any production use.

Links

Tags : LLM, Google, Gemini, Pro, Raisonnement