Gemini 3.7 Flash

2026-08-13 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 3.7 Flash, its smartest workhorse model for coding and agents, three weeks after 3.6 Flash, with an introductory price half as high.

What changes

  • Progress in debugging, incident resolution and production code (FrontierCode 43.6% vs 34.4%, DeepSWE 65.3% vs 49.0%).
  • Web development: Elo of 1588 vs 1538 on WebDev Arena.
  • Better handling of complex documents (GDP.pdf 34.0% vs 22.0%) and automation workflows (AutomationBench 30.4% vs 17.0%).
  • Agentic video understanding (support added on September 1 for 3.7, 3.6 and 3.5 Flash-Lite).
  • Becomes the model behind the Gemini Spark personal agent.

### Specifications - Context: up to 1 million tokens; output up to 64,000 tokens; text, image, audio and video inputs. - Knowledge cutoff: March 2026. - Availability: Gemini API, AI Studio, Antigravity, Gemini Enterprise Agent Platform, Gemini app. - API identifier: gemini-3.7-flash. ### Pricing | Period | Input (USD / 1M) | Output (USD / 1M) | |---|---|---| | Until 12/31/2026 (introductory) | 0.75 | 3.75 | | From 01/01/2027 | 1.50 | 7.50 | Batch and Flex: 0.375 and 1.875 (until 12/31/2026). ### Benchmarks | Benchmark (model card, scores published by the vendor) | 3.7 Flash | 3.6 Flash | Claude Sonnet 5 | GPT-5.6 Terra | |---|---|---|---|---| | DeepSWE v1.1 | 65.3% | 48.6% | 53.8% | 69.6% | | Terminal-Bench 2.1 | 85.8% | 78.0% | 80.4% | 87.4% | | GDPVal-AA v2 (Elo) | 1525 | 1422 | 1598 | 1578 | | LVBench | 85.4% | 84.2% | 68.5% | 78.9% | | OSWorld-2.0 | 47.9% | 33.8% | not published | 50.2% | | HLE-Verified | 53.6% | 51.2% | 31.0% | 51.1% | ### Key takeaways A Flash model that gets closer to more expensive models on coding and agent tasks, at a reduced introductory price until the end of 2026.

Links

Tags : LLM, Google, Gemini, Flash, Agents