Gemini 3.8 Flash

Gemini 3.8 Flash, released on 2 September 2026, is Google's most capable Flash-class model, positioned for long-horizon software engineering, autonomous agents and multi-step enterprise reasoning. It arrived three weeks after Gemini 3.7 Flash and was the third Flash release in six weeks. Technically the envelope is unchanged: a 1 048 576-token input window, 65 536 tokens of output, text, image, video, audio and PDF in, text out, and the same three thinking levels (low, medium, high — minimal returns an error). The published benchmark figure is 54.9% on HLE-Verified; for DeepSWE v1.1, Vals Finance Agent V2 and Harvey's Legal Agent Benchmark Google claims wins over 3.7 Flash and larger frontier models without printing the numbers. The interesting part is the cost. The list price is identical to 3.7 Flash — USD 0.75 per million input tokens and USD 3.75 per million output — and both rates double on 1 January 2027, so this is the same introductory window 3.7 Flash opened, not a new discount. But Google states plainly that the gains come from the model working harder: on complex tasks it takes extra reasoning steps and calls tools iteratively, and „might use more tokens to maximize performance, especially at higher effort levels”. Identical rates therefore do not mean identical bills, and the company explicitly tells cost-sensitive users to lower the effort level or stay on 3.7 Flash, which remains fully supported. A second variant, Gemini 3.8 Flash Cyber, shares the same foundation model but is not sold openly: it goes to vetted defenders through the Fairwind Program — government authorities, critical infrastructure operators and software maintainers. It reaches pass@1 of 47.2% on CWE-Bench patching (against 47.8% for the leading frontier model, at a far lower cost) and exceeds a 70% success rate on Google's internal vulnerability-discovery benchmark spanning 20 programming languages. Google reports that the Chrome Security team obtained 2.6 times more correct vulnerability patches from it than from much larger commercial models, and that its Cloud Vulnerability Research team used it to find a critical foundational flaw in under two hours.

▸News

Two flagship models launched on the same day. One promises to spend more tokens, the other to spend fewer

9/10/2026 · Google / Meta AI Research / Meta Model API docs

▸Videos

No videos yet.

▸Reviews