Google has launched Gemini 3.6 Flash and Gemini 3.5 Flash-Lite as generally available API models. Both are production-ready, support a one-million-token context window and up to 64,000 output tokens, and include thinking plus built-in tools such as Computer Use.
The pricing headline is a lower output rate for Gemini 3.6 Flash than Gemini 3.5 Flash while the input rate stays unchanged. Flash-Lite becomes the lowest-cost GA option in the Gemini 3.5 family. The live pricing table and chart above pull current rates from our dataset at build time.
See the complete Google Gemini API pricing guide, compare live rates on the Google AI pricing page, or test your workload in the token calculator. For a second provider benchmark, use the OpenAI pricing page.
What launched and the pricing impact
| Model | Default thinking | Best fit | Suggested migration from |
|---|---|---|---|
| Gemini 3.6 Flash | Medium | Agentic workflows, coding, multimodal and spatial reasoning | Gemini 3.5 Flash, Gemini 3 Flash, Gemini 3.1 Pro |
| Gemini 3.5 Flash-Lite | Minimal | High-volume extraction, parsing, analysis and autonomous subagents | Gemini 3.1 Flash-Lite, Gemini 2.5 Flash |
Google says Gemini 3.6 Flash completes multi-step workflows with fewer reasoning steps, turns, and tool calls than Gemini 3.5 Flash. That can improve effective cost beyond the rate-card reduction when a task needs fewer retries or shorter execution loops.
Gemini 3.5 Flash-Lite targets throughput-sensitive work where a lower default thinking level is useful. Buyers should still benchmark successful-task cost: a cheaper call is not cheaper if extraction errors or failed tool calls create more retries.
Gemini 3.6 Flash is a direct price-performance challenge to the previous Flash generation. Its input rate is unchanged, but its output rate is lower. Teams with output-heavy agents, coding workflows, or tool loops should see the clearest rate-card benefit.
The new Flash-Lite model is positioned differently. It prioritizes latency and volume over maximum reasoning depth, making it the more natural first test for structured parsing, document extraction, classification, and lightweight subagents.
Do not compare the models on token prices alone. Measure total input, reasoning and output tokens; tool-call count; latency; retries; and completed-task rate. Our context-window cost guide explains why long context can dominate the bill even when the per-token rate looks attractive.
Required API changes and migration steps
The migration has a code-level catch. Starting with these models, Google says temperature, top_p, and top_k are deprecated and ignored. Future Gemini model generations will return an HTTP 400 error when requests include them, so remove the parameters now rather than relying on silent behavior.
Google also requires removing prefilled model turns. Applications that previously forced an output format with a prefilled assistant response should move that instruction into system_instruction or use structured outputs.
This matters for production evaluations. If an old harness assumes temperature still controls determinism, results from the new models may be misleading even though the request succeeds. Update the request schema before comparing quality, latency, or cost.
Migration plan: Start with a shadow test against the model you use today. Keep prompts, tools, and evaluation cases fixed; remove the deprecated fields; and record successful-task cost rather than cost per request.
Use Gemini 3.6 Flash when tool use, coding, multimodal input, or complex workflows drive the decision. Start with Gemini 3.5 Flash-Lite when throughput and predictable structured output matter more than deep reasoning.
Pin the exact model identifier during rollout. Google has also made Gemini 3.6 Flash the new default for its Antigravity managed agent, but production applications should not depend on an implicit default when behavior and token use affect the bill.
Bottom line
Gemini 3.6 Flash is the stronger migration target for teams that want better agentic performance and lower output pricing than the previous Flash model. Gemini 3.5 Flash-Lite is the volume route. Both are GA, but both require a request-schema cleanup that should happen before any cost or quality benchmark.
Sources: Google’s official latest Gemini models guide, Gemini API pricing, and AI Pricing Guru’s live pricing dataset.