Gemini 3.8 Flash API
Choose Gemini 3.8 Flash
Google’s newest Flash-tier workhorse for coding, agentic workflows, knowledge work, and multimodal reasoning, with higher reported accuracy than Gemini 3.7 Flash and a 1M-token context window.
Gemini 3.8 Flash
Google Flash-tier workhorse model
gemini-3.8-flashCoding, multi-step agents, and document or chart analysis—the workloads Google highlights. EvoLink has not published task-level cost or acceptance results for these cases.
Gemini 3.8 Flash pricing
Estimate a request with the interactive pricing calculator and compare EvoLink with Google’s current introductory rate.
Token calculator
Enter the token mix for one request.Estimated request cost
Gemini 3.8 FlashEstimated minimum charge: 0.01 credits per request.
EvoLink vs Google direct
Same token mix, default group price.Budget guide
Approximate requests using the current token mix.For quick testing
For regular development
For production evaluation
Model pricing
| Model | Context | Input tokens | Cache read tokens | Output tokens |
|---|---|---|---|---|
Gemini 3.8 Flashgemini-3.8-flash | All context sizes | $0.675 / 1M-55% 45.9 cr / 1M$1.500Google intro price | $0.068 / 1M-55% 4.6 cr / 1M$0.150Google intro price | $3.375 / 1M-55% 229.5 cr / 1M$7.500Google intro price |
Gemini 3.8 Flash
All context sizesUSD and credits are shown per 1M tokens. Live backend pricing takes priority over these frozen fallback rates.
Gemini 3.8 Flash API for coding and agent workflows
Call Google’s newest Flash-tier workhorse through EvoLink’s unified API. Google reports higher accuracy than Gemini 3.7 Flash, alongside higher token consumption. Gemini 3.8 Flash supports a 1,048,576-token context window, prompt caching, structured output, and tool use — at Google’s introductory $0.75 / $3.75 per million input / output tokens through December 31, 2026.
How to evaluate Gemini 3.8 Flash in a production model stack
Google positions Gemini 3.8 Flash as a higher-accuracy workhorse for coding, knowledge work, and multimodal tasks, while documenting higher token consumption than Gemini 3.7 Flash. Google does not publish workload-specific total-cost results, so production fit must be established with your own controlled evaluation.
Coding and refactoring evaluation
Gemini 3.8 Flash targets everyday coding, prototyping, and full-stack refactoring with stronger reported accuracy than Gemini 3.7 Flash, but potentially higher token use. Measure accepted patches, total tokens, and review time—not isolated snippet quality.
Agentic workflows and orchestration
Google highlights multi-step orchestration, tool selection, structured output, and code execution. EvoLink has not measured task completion or total cost for these workflows; preserve complete assistant messages, tool-call IDs, arguments, and tool results when testing them.
Knowledge work and multimodal reasoning
Use it for document analysis, chart interpretation, and multi-element web layout generation across a 1M-token context. Retrieval and document structure still matter: a 1M window does not make irrelevant context useful.
When another route is the better choice
Gemini 3.5 Flash-Lite has a lower published token rate for high-volume classification and extraction. Do not assign 3.8 Flash based on an assumed task-cost advantage; compare it with your existing route under the same acceptance criteria.
What Google confirms—and what has not been measured by EvoLink
Google released Gemini 3.8 Flash on 2026-09-02 and published model-level benchmark and token-consumption statements. EvoLink has not published a controlled 3.8-versus-3.7 workload study, so this page does not claim fewer retries, shorter agent loops, or lower task-level cost.
Same introductory token rate, higher official benchmark scores
Google lists the same introductory input, output, and cache-read rates for 3.8 Flash and 3.7 Flash, and reports higher 3.8 scores on highlighted coding and agentic benchmarks. Google also documents higher token consumption; those facts do not establish a lower bill for a production task.
Official benchmarks are not a cross-provider production ranking
Google’s release results describe selected benchmarks and do not prove that 3.8 Flash beats every Pro-tier or third-party model on your workload. This page therefore makes no cross-provider winner claim.
Higher token consumption is an official caveat
Google explicitly says 3.8 Flash improves accuracy while using more tokens than 3.7 Flash. The published token rates alone cannot determine which model costs less for a completed task.
It is a closed, API-only Google model
Gemini 3.8 Flash has no open weights and cannot be self-hosted. Access is through the Gemini API and platforms such as Google AI Studio, Antigravity, and unified gateways like EvoLink—so route by cost and reliability, not by local deployment.
Documented capabilities and their evaluation boundaries
Google documents a large context window, higher accuracy, and prompt caching. It does not publish EvoLink workload results showing that these features reduce completed-task cost; context capacity alone is not evidence of better application outcomes.
A 1M-token workspace, not a target to fill
The 1,048,576-token window can keep related code, specifications, and prior tool results in one request. Whether that improves an application outcome still requires testing; retrieval and context compaction matter because irrelevant input also consumes tokens.
Higher accuracy with higher token use
Google documents higher accuracy alongside higher token consumption than Gemini 3.7 Flash. Track total cost per accepted task to determine whether better results outweigh the additional input, output, and reasoning tokens.
Prompt caching pays off when prefixes stay stable
Repository instructions, system prompts, reference material, and tool schemas create the strongest cache opportunity when their ordering stays consistent. Frequent model or prompt-structure changes can force the long prefix to be processed again.
What to verify before routing production traffic to Gemini 3.8 Flash
A suitable workload can still fail because the integration uses the wrong identifier, sends unsupported parameters, or drops agent state between turns. Verify the request surface and conversation contract before evaluating model quality.
Use the exact model ID gemini-3.8-flash
Send model "gemini-3.8-flash" (with dots) on the EvoLink API route. The dashed form gemini-3-8-flash is only the page URL, not the API model parameter.
Use OpenAI Chat Completions or the Gemini native API
EvoLink exposes Gemini 3.8 Flash through OpenAI-compatible /v1/chat/completions and the Gemini native generateContent endpoint. Keep the same EvoLink API key for either protocol.
Mind the breaking parameter changes
For EvoLink native requests, Gemini 3.x uses generationConfig.thinkingConfig.thinkingLevel; thinkingBudget is for Gemini 2.5 and the two controls are mutually exclusive. EvoLink documents that custom temperature and topP values do not affect Gemini 3.x output, topK is ignored, and out-of-range temperature or topP values return 400. The final conversation turn must not use role model.
Replay complete assistant and tool state
Multi-turn agents should retain complete assistant messages, tool-call IDs, arguments, and tool results. Keeping only the final text breaks state continuity and can make later steps fail even when the context window is large enough.
No task-level cost winner has been established
The verified facts are that 3.8 Flash and 3.7 Flash share the same Google introductory token rates, while Google says 3.8 uses more tokens and achieves higher accuracy. EvoLink has not published controlled results for retries, accepted deliverables, tool calls, elapsed time, or human review.
These are evaluation inputs, not observed Gemini 3.8 Flash outcomes. Until the same production samples are tested under the same settings, do not claim that 3.8 lowers cost per accepted task; choose 3.7 or Flash-Lite when their documented price and compute profile already match the workload.
Compare leading long-context models after workload testing
EvoLinkMeasure task accuracy and accepted-result cost against Gemini 3.8 Flash’s higher token consumption. Then compare price, context, caching, and workload fit.
| Model | Gemini 3.8 Flash | Gemini 3.5 Flash Lite | Gemini 3.1 Pro |
|---|---|---|---|
| Input / output | $0.675 / $3.375 | $0.3 / $2.5 | $1.68 / $10.08 |
| Context | 1M | 1M | 1M |
| Caching | Automatic cache reads | Context cache | Context cache |
| Best for | Coding, multi-step agents, and document or chart analysis—the workloads Google highlights. EvoLink has not published task-level cost or acceptance results for these cases. | A lower published-rate route to compare for classification, extraction, and high-throughput tasks. | A Pro-tier comparison route for tasks that require deeper reasoning than the Flash line provides. |
Other Gemini models on EvoLink
Gemini 3.5 Flash Lite
A lower published-rate route to compare for classification, extraction, and high-throughput subagents.
View model
Gemini 3.1 Flash Lite
A lightweight 3.1 route to compare for cost-sensitive, high-volume workloads.
View modelOther text models
GPT-5.6
A tiered OpenAI model family for capability, latency, and cost-routing comparisons.
View model
Claude Opus 4.8
An Anthropic model to compare on long-running coding agents and judgment-heavy review.
View model
Related reading for production teams
Gemini 3.8 Flash vs Gemini 3.7 Flash
Compare accuracy, token consumption, introductory pricing, and cost per accepted production task before changing your default.
Read guide
How to use Gemini 3.8 Flash API
Make the first request, handle Gemini 3 migration rules, and add telemetry, canary, and rollback controls.
Read guide
Compare the Gemini API family
See how 3.8 Flash, 3.5 Flash-Lite, 3.7 Flash and 3.1 Pro differ on price, context, and workload fit before you route.
Read guide
Gemini 3.5 Flash-Lite API
A lower published-rate 3.5-class route to compare with 3.8 Flash for classification, extraction, and high-throughput subagents.
Read guide
Gemini 3.7 Flash API
The previous-generation Flash model at the same price — compare it before moving production traffic to 3.8 Flash.
Read guideGemini 3.1 Pro API
Step up to the Pro tier when a task needs deeper reasoning than the Flash line can deliver.
Read guideGemini 3.8 Flash API FAQ
Is the Gemini 3.8 Flash API available through EvoLink?
Yes. Gemini 3.8 Flash is listed as a production model. The page uses backend pricing when returned and a current Google introductory-rate fallback for comparison.
What model ID should I use for the Gemini 3.8 Flash API?
Send model "gemini-3.8-flash" (with dots). The dashed gemini-3-8-flash is only the page URL, not the API model parameter.
Which protocols and SDKs work with Gemini 3.8 Flash?
EvoLink documents OpenAI-compatible Chat Completions and Gemini native generateContent on direct.evolink.ai with the same API key, and gemini-3.8-flash is listed in the model enum of both endpoints. Still confirm the live route in your account.
Does the Gemini 3.8 Flash API really support the full 1M context?
Yes — the route records 1,048,576 input tokens per Google’s documentation. Smaller limits you may see usually come from a particular Gemini product surface or client configuration, not the API model itself.
How should I use the Gemini 3.8 Flash 1M-token context window?
Keep related code, documents, and tool results together, but use retrieval, stable cached prefixes, and context compaction instead of filling the window by default.
How is Gemini 3.8 Flash different from Gemini 3.7 Flash and 3.5 Flash-Lite?
Google lists the same introductory token rates for 3.8 and 3.7, reports higher 3.8 scores on selected coding and agentic benchmarks, and documents higher 3.8 token consumption. Google lists lower token rates for 3.5 Flash-Lite; workload outcomes still require testing.
Is Gemini 3.8 Flash actually cheaper to run if it uses more tokens?
Per-token rates match Gemini 3.7 Flash, but Google documents higher token consumption alongside higher accuracy. Compare total cost per accepted task on your own workload rather than assuming a lower per-task cost.
What parameter rules should I handle when migrating to Gemini 3.8 Flash?
For EvoLink native requests, Gemini 3.x uses generationConfig.thinkingConfig.thinkingLevel; thinkingBudget is for Gemini 2.5 and cannot be combined with thinkingLevel. EvoLink documents that custom temperature and topP do not affect Gemini 3.x output, topK is ignored, and out-of-range temperature or topP values return 400. The final turn must not use role model.
What happened to the minimal thinking level in Gemini 3.8 Flash?
Google documents low, medium (default), and high for Gemini 3.8 Flash; minimal is unsupported. EvoLink’s native API reference states that an unsupported minimal is automatically downgraded to low, so migrated requests do not fail, but set low explicitly for predictable control. Gemini 3.5 Flash-Lite has lower published token rates and can remain a comparison route for high-volume classification.
How are Gemini 3.8 Flash thinking tokens billed?
Thinking tokens are billed at the output rate under Google’s pricing, so verbose reasoning can make bills noticeably larger than the visible answer. Set task-appropriate thinking levels and monitor reasoning plus final-answer tokens.
Can Gemini 3.8 Flash analyze video and audio?
Yes. Google documents text, image, video, audio, and PDF input with text output. That makes video and audio understanding valid evaluation cases; it does not support image, audio, or Live API generation.
Is there a Gemini 3.7 Pro?
No. As of the 3.8 Flash release, Google’s newest Pro-line model remains gemini-3.1-pro-preview, and no 3.5 or 3.7 Pro has been announced on official channels.
How long does the Gemini 3.8 Flash introductory price last?
Google lists $0.75 input and $3.75 output per million tokens as an introductory price through December 31, 2026, with standard rates of $1.50 and $7.50 from January 1, 2027. See the pricing section on this page for the current EvoLink rate.
Is Gemini 3.8 Flash more reliable than 3.7 Flash at following instructions?
Google reports higher overall accuracy and higher token consumption, but EvoLink has not published a controlled instruction-following comparison. Test representative tasks before promoting it to production.
What should a production Gemini 3.8 Flash evaluation measure?
No EvoLink comparison result is published yet. Evaluation inputs should include first-pass success, accepted deliverables, retries, output tokens, cache hits, valid tool calls, time to accepted result, human correction, and fallback rate.