Gemini 3.8 Flash

2026-09-02 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 3.8 Flash, its most intelligent Flash model, better at code, agents and multi-step reasoning at the same price as 3.7 Flash, along with the cybersecurity variant 3.8 Flash Cyber.

What changes

  • Third Flash version in six weeks, more efficient on long tasks: DeepSWE v1.1 at 73.7%.
  • The model “works harder”: more reasoning steps and tool calls, at the cost of sometimes higher token usage.
  • Adjustable effort levels to balance quality, cost and latency.
  • 3.8 Flash Cyber: vulnerability detection and automatic fixing, reserved for trusted defenders through the Fairwind program.
  • 3.7 Flash remains supported for uses where efficiency comes first.

### 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 app, Gemini API, AI Studio, Antigravity, AI Mode, Gemini Enterprise Agent Platform. - API identifier: gemini-3.8-flash. ### Pricing | Period | Input (USD / 1M) | Output (USD / 1M) | |---|---|---| | Until December 31, 2026 (introductory) | 0.75 | 3.75 | | From January 1, 2027 | 1.50 | 7.50 | Batch and Flex: 0.375 and 1.875 (until December 31, 2026). ### Benchmarks | Benchmark (model card, scores published by the vendor) | 3.8 Flash | 3.7 Flash | Claude Opus 5 | GPT-5.6 Sol | |---|---|---|---|---| | DeepSWE v1.1 | 73.7% | 65.3% | 74.0% | 72.7% | | Terminal-Bench 2.1 | 89.4% | 85.8% | 89.1% | 88.8% | | GDPVal-AA v2 (Elo) | 1545 | 1482 | 1824 | 1710 | | Vals Finance Agent v2 | 61.4% | 59.0% | 58.6% | 53.8% | | OSWorld-2.0 | 59.0% | 50.6% | 75.4% | 62.6% | | HLE-Verified | 54.9% | 53.6% | 54.4% | 54.5% | ### Key takeaways A cost-contained working model that matches or approaches the more expensive frontier models on several code and financial analysis benchmarks; the price doubles on January 1, 2027.

Links

Tags : LLM, Google, Gemini, Flash, Agents