Gemini 3.6 Flash vs 3.5 Flash-Lite: API Migration Guide

Short summary

Google released Gemini 3.6 Flash and 3.5 Flash-Lite as GA models on July 21, 2026, each targeting different production workloads. Flash handles coding, multimodal reasoning, and agent workflows; Flash-Lite serves extraction, classification, and routing at lower cost. Migration requires removing deprecated temperature/top_p/top_k fields, replacing thinking_budget with thinking_level, and eliminating prefilled model turns or facing HTTP 400 errors.

• •3.6 Flash for coding and agent workflows ($1.50/1M input); Flash-Lite for extraction and routing ($0.30/1M input)
• •Migration breaks: deprecated temperature/top_p/top_k, prefilled model turns rejected, thinking_budget replaced by thinking_level
• •Route by workload—test both models on same fixtures before changing production defaults

Generated with AI, which can make mistakes.