Gemini 3.8 Flash — Longer-Horizon Coding At Flash Cost

September 3, 2026

Google's Gemini 3.8 Flash is now in the model selector for long-horizon software engineering, autonomous agent work, and difficult multi-step analysis. You get a 1M-token context window, multimodal input, tools, and medium-effort reasoning at Workhorse 3×.

What you can do

• Pin Gemini 3.8 Flash for work that needs more persistence than an everyday fast model — large refactors, iterative tool loops, and specialized research that has to hold together across many steps.
• Bring text, images, audio, video, or PDFs into the same turn. The model accepts up to 1,048,576 input tokens and can return up to 65,536 output tokens.
• Use tools and structured output. Gemini 3.8 Flash can call Alfrada OS tools, act as a Swarm worker, and return schema-shaped results for workflows that need dependable handoffs.
• Tune the depth with Effort. Medium reasoning is the default; lower effort reduces token use for faster iterations, while higher effort gives complex coding and analysis more room.
• Reuse long prompt prefixes more cheaply. Gemini 3.8 Flash supports implicit and explicit context caching; cached input is billed at 10% of the normal input rate.

Where this shows up

You have a repository-wide migration that needs repeated searches, edits, tests, and corrections. Pin Gemini 3.8 Flash so the same model can keep the plan coherent while it works through the long tool loop.

You are combining a PDF brief, product screenshots, and interview audio into one decision memo. Gemini 3.8 Flash can reason across those inputs in a single large context and return the findings in a structured format.

You want a second model to pressure-test a Swarm deliverable without paying for a frontier SOTA worker. Add Gemini 3.8 Flash as the coding or QA worker at Workhorse 3×.

Try it

• "Use Gemini 3.8 Flash to migrate this repository to the new API, run the focused tests after each stage, and keep fixing issues until the suite passes."
• "Compare this contract PDF with the policy document and meeting recording, then return every conflict as structured JSON with supporting evidence."
• "Add Gemini 3.8 Flash as the QA worker in this Swarm and have it challenge the implementation plan, edge cases, and test coverage."

Heads up

• Auto can pick Gemini 3.8 Flash on every plan. At quality 60 it sits exactly on the paid Auto floor (60), so Pro, Pro Max, and Black Auto may route to it as a Workhorse — unlike Gemini 3.7 Flash, which sat under the floor at 59 and was Community-Auto only. Pin it when you want it every turn.
• Gemini 3.7 Flash is retired. Existing sessions pinned to 3.7 — and older 3.5 or 3.6 Flash sessions — roll forward to Gemini 3.8 Flash automatically. Gemini 3.7 also leaves Swarm and worker selection.
• Introductory provider pricing runs through December 31, 2026: $0.75 per 1M input tokens and $3.75 per 1M output tokens. Google says this changes on January 1, 2027 to $1.50 input / $7.50 output. Alfrada currently bills the model at Workhorse 3×.
• Pricing and model specifications come from Google's Gemini 3.8 announcement and official model documentation.