Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- gemini-3.8-flash is GA for long-running software engineering, autonomous agents and complex enterprise workflows. 3.7 Flash remains fully supported, with no announced shutdown date.
- Through December 31, 2026, 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens. From January 1, 2027, those prices become $1.50 and $7.50.
- It supports 1,048,576 input tokens, 65,536 output tokens and low, medium and high thinking levels. minimal produces a validation error.
- Migration requires more than changing the model ID: remove temperature, top_p, top_k, candidate_count and prefilled model turns, and replace thinking_budget with thinking_level.
- Run 3.7 and 3.8 on the same real tasks and compare quality, latency, tokens, tool failures and cost per task. Keep 3.7 without a clear gain; a version number is not acceptance evidence.
3.8 is GA, but 3.7 does not need immediate retirement
If your main work involves complex code changes, multi-step tool calls or long-context agents, gemini-3.8-flash merits a staged test. Google's GA designation and positioning for long-running software engineering, autonomous agents and complex enterprise workflows establish its release stage. They do not prove it outperforms 3.7 for your repository, region, quotas or latency targets.
Google's deprecation page says 3.7 Flash remains fully supported with no shutdown date. Do not upgrade simple classification, fixed-format extraction or extremely latency-sensitive tasks merely for the version number. Compare identical real samples; retain 3.7 if gains do not cover token usage, thinking time and migration costs.
Understanding temporary 3.8 prices and 2027 costs
The official temporary 3.8 Flash price through December 31, 2026 is $0.75 per million input tokens and $3.75 per million output tokens. From January 1, 2027, standard prices are $1.50 input and $7.50 output. Both double, so temporary pricing should not determine a long-term budget.
Track at least input, caching, output, tool calls and failed retries separately. High thinking levels can increase reasoning and output consumption; long contexts increase the cost of a failed request. Calculate procurement and capacity decisions at both current temporary and 2027 standard prices.
A 1M context limit does not mean you should fill it
The 3.8 model page lists a 1,048,576-token input limit and 65,536-token output limit, with text, image, video, audio and PDF input and text output. It supports caching, code execution, function calling, search, structured output and thinking. Computer Use remains Preview; Live API, image generation and audio generation are unsupported.
Requests near the limit increase latency, cost and diagnostic difficulty. Filter and retrieve files for repositories, and summarize documents in layers. Include relevant evidence and record truncation, cache hits and tool results.
What to change when migrating from 3.7 or older models
Changing the model identifier to gemini-3.8-flash is only the first step. Google's migration checklist requires removing temperature, top_p, top_k and candidate_count, replacing thinking_budget with thinking_level, and removing prefilled model turns. The final user turn must also contain nonempty text.
Check relationships among call_id, name, function responses and thought signatures. Run both 3.7 and 3.8 on identical inputs in a test environment and record actual transmitted parameters. Do not silently discard incompatible parameters and switch all traffic immediately.
Choosing among the three thinking levels
medium is the default for 3.8 Flash and a starting point for complex coding and agent tasks. low better suits incident response, live chat, drafts and quick analysis. high targets difficult reasoning, mathematics and long tool chains, usually with more tokens and time. minimal is unsupported and returns a validation error.
Do not fix one level across the product. Route low-risk quick tasks to low, default workflows to medium, and only difficult tasks with demonstrated benchmark gains to high. Set budgets and timeouts for expensive requests.
Staged release and rollback checks
Select real samples with successes, failures and edge cases, and record a 3.7 baseline. Compare accuracy, human rework, time to first token, total time, input/output tokens, tool success, format compliance and cost per task. One demo, the model's self-reported version or an official benchmark cannot replace your own acceptance tests.
Enable 3.8 for a small share of requests first, log task type and model ID, and retain a per-request switch back to 3.7. Roll back immediately for formatting errors, tool loops, cost or latency overruns, or critical quality regressions. Expand gradually after stability is demonstrated.
Distinguish model configuration errors from network failures
If 3.8 is absent or returns model not found, first identify whether you are using the API, AI Studio or Gemini app, then check model ID, SDK version, project and regional permissions. Do not infer routing by asking the model its version. Separate 429, 503 and 403 by quota, capacity and permission using the Gemini API 429 and 503 guide. Consult the proxy connection failure checklist only with clear DNS, connection-timeout, TLS or proxy-authentication evidence.
For stable access to overseas API documentation and consoles, teams can visit the PuppyIP website to learn about fixed network exits. A network environment can improve the connection path, but cannot change official prices, quotas, model permissions or regional restrictions.
Sources
Frequently Asked Questions
Is Gemini 3.8 Flash GA?
Yes. Google marked gemini-3.8-flash GA on September 2, 2026. Actual project access still depends on the API model list, AI Studio, project permissions and regional availability.
What does the Gemini 3.8 Flash API cost?
Through December 31, 2026, temporary prices are $0.75 per million input tokens and $3.75 per million output tokens. From January 1, 2027, standard prices are $1.50 and $7.50.
Does 3.8 GA immediately retire 3.7 Flash?
No. Google's deprecation page currently lists 3.7 Flash as fully supported without a shutdown date. Decide migration through real-task acceptance; do not mistake a new release for retirement of the old version.
Is changing the model name to gemini-3.8-flash enough?
No. Remove old sampling parameters, candidate_count and prefilled model turns, switch to thinking_level, and check user turns, thought signatures and tool-call fields.
Should thinking_level be low, medium or high?
Use medium as a default baseline. Test low for simple latency-sensitive tasks, and use high only where difficult tasks show quality gains worth the additional cost. minimal is unsupported.
Should a 1M context hold the whole repository?
No. The context limit is not a recommended amount. Retrieve and filter relevant files first to avoid irrelevant content increasing latency, cost and reasoning noise.
Can a proxy fix missing Gemini 3.8 access or a 429?
No. A proxy changes only the network path, not model IDs, project quotas, subscription eligibility, staged access or regional restrictions. Identify the failure layer from the product entry point and error evidence first.