Gemini 3.8 Flash

Google's stable Gemini 3.8 Flash model for coding, agentic, and complex multistep work. These notes describe model behavior, not Gemini CLI behavior.

• State the outcome, constraints, relevant project context, and completion checks.
• Break long-horizon work into verifiable deliverables and provide the tools or files needed to verify them; the model may take iterative reasoning and verification steps on complex work.
• When the calling surface exposes thinking levels, use medium for most complex coding or agentic work, high for difficult multi-step reasoning, and low for latency-sensitive tasks.
• Use the exact stable model identifier when reproducibility matters more than automatically following a moving alias.

• Assuming a latest alias remains pinned to one model version.
• Treating broad model-family claims as guarantees for a particular workload.
• Treating iterative tool use or verification as a substitute for explicit acceptance criteria and validation commands.

The minimal thinking level is unsupported for Gemini 3.8 Flash. Availability, quotas, tool capabilities, and behavior can vary by product and access tier; validate important workflows against the exact endpoint and harness in use.