Google made Gemini 3.8 Flash generally available on September 2, 2026, shipping it under the model ID gemini-3.8-flash as its most intelligent Flash model, aimed at long-horizon software engineering, autonomous agents and complex enterprise workflows. The most interesting difference between it and its two obvious rivals, OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1, is not a leaderboard position. It is what you are allowed to put into the prompt in the first place.
Gemini 3.8 Flash takes text, images, video, audio and PDFs as input and returns text, with a 1M-token input context window and up to 64k output tokens. GPT-6 Astra is the bigger container on paper, at a 1,050,000-token context window and 128,000 max output tokens, but OpenAI’s model page lists audio and video as not supported. Claude Fable 5.1 offers a 1M-token context and 128K max output, and Anthropic’s documentation is blunt that all current Claude models support text and image input. Neither rival reads a video file. You feed them extracted frames or a transcript you produced yourself, or you feed them nothing.
The feature that turns raw video support into something useful arrived one day earlier. On September 1, 2026 Google released agentic video understanding, and the Gemini API documentation lists Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash and 3.5 Flash Lite as the models that support it. Instead of chewing through footage at a fixed frame rate, the model decides what to watch, at what speed, and through which modality, pulling frames, audio or transcript on demand through internal tools. Google’s claim for the approach is specific: it can “reduce analysis costs by up to 66% and token consumption by up to 88%, while improving accuracy by up to 7%”.
The practical boundaries Google publishes:
- Models with a 1M-token context window handle videos up to 3 hours long by default at low media resolution, or up to 1 hour long at high media resolution.
- The Files API accepts videos up to 20GB on paid tiers and 2GB on the free tier.
- Agentic processing “uses standard Gemini API token pricing with no additional feature fee”.
- It is reachable through Google AI Studio, the Gemini Enterprise Agent Platform and the standard Gemini API via an “agentic” processing setting.
- Agentic mode can raise time to first token on short clips, so Google recommends static mode for videos under 5 minutes.
Google also lists benchmark results for the model: over 70% on DeepSWE v1.1, 61.4% on Vals Finance Agent v2 and 54.9% on HLE-Verified. Google said the Gemini app rollout would follow, with YouTube’s “Ask YouTube” feature arriving after that.
None of this makes the competition irrelevant. GPT-6 Astra and Claude Fable 5.1 are both built for hard reasoning and long-running agentic work, and both are priced identically at $10 per million input tokens and $50 per million output tokens. But if your source material is an hour of screen recording, a lecture or a support call, only one of the three will look at it.
