- Published: September 3, 2026
- Last verified: September 3, 2026
- Current status: Active;
gemini-3.8-flash is listed in the LaoZhang API console, with token-group access and live pricing shown there
gemini-3.8-flash is now available on LaoZhang API with usage-based Chat billing, launching at $0.75 per 1M input tokens and $3.75 per 1M output (completion) tokens in the console on September 3, 2026. Before migrating, confirm that the target token group can see the model and run a small acceptance test with representative prompts.
Key facts
The token limits and thinking levels above come from Google’s model specifications. If you use tool calls, structured output, multimodal input, streaming, or thinking parameters, send one small request for each with your target token before you switch production traffic.
Who should pay attention
- Developers selecting a Flash model for long-horizon software engineering, code refactoring, or complex agent workflows;
- Teams using
gemini-3.6-flash or an earlier Flash model and planning matched-workload tests;
- Production applications reassessing input cost, output quality, latency, and total token cost.
Image generation, audio generation, and video generation workloads that do not use Gemini text models are not affected. Stable production workloads do not need to migrate immediately; this announcement adds a model and does not retire an older one.
How to call Gemini 3.8 Flash
First confirm that the target token group can see gemini-3.8-flash on the console model and pricing page, then send a minimal text request. Replace the placeholder with your LaoZhang API key:
A basic acceptance test should confirm that the response contains usable text, the returned model matches the request, call logs record input and output tokens, and the charge matches current console pricing. Add a separate test for each required capability—tool calls, structured output, multimodal input, or streaming—instead of treating one HTTP 200 response as proof of full compatibility.
Migration notes
Google’s migration guide describes these thinking levels for Gemini 3.8 Flash:
- The default thinking level is
medium.
- Supported levels are
low, medium, and high.
minimal returns an error.
Native Gemini API migrations also need to check deprecated sampling parameters, thinking_budget, candidate_count, and turn-format rules.
Those native rules should not be assumed to match LaoZhang API’s OpenAI-compatible protocol. For a production migration:
- Replace the model ID only in a test environment and keep a verified fallback model.
- Compare answer quality, time to first token, total latency, and token usage on representative requests.
- Test tool calls, JSON output, image or file input, and streaming separately.
- Reconcile the actual charge in call logs before increasing traffic in stages.
Frequently asked questions
What is the correct Gemini 3.8 Flash model ID?
Use gemini-3.8-flash. Do not append -preview, -thinking, or -nothinking; use a compatibility alias only if the console explicitly lists it.
Where do I find the current price?
The launch price from September 3, 2026 is at the top of this page. Pricing can change with routes or promotions, so check the model and pricing catalog and the console before a large workload, and use call logs as the billing record.
Is Gemini 3.8 Flash a production model?
Yes. Google marks gemini-3.8-flash as GA with a stable model ID. Access for a specific LaoZhang API token group must still be confirmed in the console.
Can I migrate by changing only the model ID?
Changing the model ID is a useful first step, but it does not prove complete compatibility. Before production migration, test prompt behavior, thinking, tool calls, structured output, streaming, token usage, and latency.
Must I stop using an older Gemini Flash model?
No. This is a model-addition announcement, not a retirement notice. Continue using an older model when its quality, latency, and cost meet your requirements, and schedule a matched-workload comparison first.
