Gemini 3.8 Flash
Google DeepMind · Multimodal model
Gemini 3.8 Flash is Google's current mid-tier 'Flash' model, built on 3.7 Flash with better coding and agent performance. It takes text, images, audio, video and PDFs with a 1M-token window. Google now auto-routes requests for 3.7 Flash and 3.5…
Current
ReleasedSep 2, 2026
Context window1,048,576 input tokens; 65,536 output tokens
Input → outputtext, image, video, audio, pdf → text
API price$0.75 / 1M input tokens and $3.75 / 1M output tokens (incl. thinking) through Dec 31, 2026; $1.50 input / $7.50 output from Jan 1, 2027. Free tier available.
Consumer access—
Best for
Fast, lower-cost coding and agentic workloads that still need strong reasoning.
Overview
Gemini 3.8 Flash is Google's current mid-tier 'Flash' model, built on 3.7 Flash with better coding and agent performance. It takes text, images, audio, video and PDFs with a 1M-token window. Google now auto-routes requests for 3.7 Flash and 3.5 Flash to it.
Strengths and limits
- Large gains over 3.7 Flash on agentic coding (DeepSWE, Terminal-Bench)
- Available on the free API tier
- Supports caching, search grounding, function calling and computer use (preview)
- Text output only
- Introductory pricing doubles from Jan 1, 2027
Published benchmark scores
DeepSWE v1.173.7%
Source: deepmind.google
Terminal-Bench 2.189.4%
Source: deepmind.google
OSWorld-2.0 (partial score)59.0%
Source: deepmind.google
HLE-Verified54.9%
Source: deepmind.google
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →