Access Google Gemini 3.8 Flash—also searched as Gemini 3.8— through EvoLink's unified chat API. Test coding, multi-step agents, tool use before integrating.
Google·Text Generation·Available
from $0.675 / 1M input tokens$1.500 official priceLimited time-55%
Long-context codingConfigurable thinkingPrompt cachingChat + Gemini API
Production routeThis rate reflects platform-side availability — only confirmed server errors (HTTP 500 / empty response) count as failures. User-side issues (content moderation, invalid params, cancellation) plus rate limits, timeouts and auth errors are excluded. Before real traffic arrives, empty buckets may display as available.Live
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
Selected
Model ID
gemini-3.8-flash
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.
Input
$0.675 / 1M-55%
45.9 cr / 1M$1.500official price
Cache read
$0.068 / 1M-55%
4.6 cr / 1M$0.150official price
Output
$3.375 / 1M-55%
229.5 cr / 1M$7.500official price
All rates are per 1M tokens, shown in USD and credits, and reflect your account's current pricing.
Gemini 3.8 Flash pricing
Estimate what one Gemini 3.8 Flash request costs before you integrate. The calculator uses your account's current rates, with official pricing as a reference.
Limited-time offer — the rates below reflect Google’s introductory pricing and run through December 31, 2026 (UTC+8). From January 1, 2027, our usual discount applies to Google’s standard price ($1.500 input / $7.500 output per million tokens).
Request calculator
Enter the token mix for one request and the number of Google Search queries.
Estimated request cost
Gemini 3.8 Flash
USD$0.0018
Credits0.1158
Official estimate $0.0038 · save $0.0021 (55%)
Input tokens0.0459 cr
Cache read tokens0.001 cr
Output tokens0.0689 cr
Google Search0 cr
Minimum charge: 0.01 credits per request.
Budget guide
Approximate requests using the current token mix and search queries.
Gemini 3.x bills each nonempty search query, even if no sources are returned. One request may run multiple queries. Search fees are added separately after the token minimum charge.
Google Search$0.014/ query·0.952 cr / query
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 $0.750 / $3.750 per million input / output tokens through December 31, 2026.
Gemini 3.8 Flash is served on EvoLink under the model ID gemini-3.8-flash through Chat Completions · Gemini native API, with the same API key and balance you use for every other model. It offers a 1.05M context window and up to 65.5K output tokens, plus coding, multi-step agents, tool use.
Gemini 3.8 Flash specs and capabilities
Numbers come from the EvoLink route configuration; capabilities are what the API exposes today.
Context window
1.05M tokens
Max output
65.5K tokens
Input
Text + images
Output
Text · JSON (structured output) · tool calls
Reasoning
Configurable reasoning effort
Tool use
Function calling with multi-step tool sequences
Prompt caching
Automatic cache reads at a lower rate
Server-side tools
Google Search, billed per search query
Protocols
Chat Completions · Gemini native API
Model ID
gemini-3.8-flash
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.
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.
Two ways to use Gemini 3.8 Flash: EvoLink API or Agent
Use the EvoLink API for product backends and batch jobs, or call Gemini 3.8 Flash from Codex, Claude, or Gemini for coding and analysis workflows. Both paths share the same EvoLink API key, balance, model ID, and request history.
Option 1
Integrate with the EvoLink API
Best for: product backends, batch jobs, automated pipelines
Send Gemini native generateContent requests to EvoLink and control the model ID, system instruction, output budget, tools, and structured output.
1Create an EvoLink API key in the console
2Call the generateContent endpoint on the EvoLink base URL with the model ID shown above in the request path
3Send one representative request and read usageMetadata for input, cached, and output tokens
4Set maxOutputTokens and retries per task; keep function-call parts and their results across turns
Best for: coding, review, and analysis tasks in Codex, Claude, and Gemini
Give the Agent the task, the inputs to include, and the acceptance criteria. It assembles the request, calls Gemini 3.8 Flash through EvoLink, and returns the answer with token usage.
1Set EVOLINK_API_KEY in your local environment; never put it in code or a prompt
2Describe the task, the inputs to include, and the expected output format
3Ask the Agent to call Gemini 3.8 Flash through EvoLink and show the request before sending
4Let the Agent report the answer, token usage, and any error body
Gemini 3.8 Flash API code example and error handling
This example shows the shortest runnable request: a Gemini native generateContent call with a system instruction, a user message, and an output budget. Open the API tab for the complete parameter and response reference.
# Gemini native API
curl -X POST "https://direct.evolink.ai/v1beta/models/gemini-3.8-flash:generateContent" \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"systemInstruction": { "parts": [{ "text": "You are a precise assistant. Answer in JSON when asked." }] },
"contents": [
{ "role": "user", "parts": [{ "text": "Classify the sentiment of each review below and return a JSON array:\n<reviews>" }] }
],
"generationConfig": {
"maxOutputTokens": 1024,
"temperature": 0.1
}
}'
# For streaming, call streamGenerateContent?alt=sse with the same body.
# The response includes candidates[0].content.parts and usageMetadata
# (prompt, cached, thinking, and output token counts).
Invalid request or unsupported parameter
Check the model ID, the contents array, and parameter ranges against the API reference; remove fields this route does not support.
Authentication or balance issue
Check the Authorization bearer token and confirm the available balance in the console.
Context length exceeded
Prompt tokens exceed the Gemini 3.8 Flash context window. Trim or retrieve only the relevant evidence and reuse cached prefixes.
Rate limited (429)
Back off and retry with jitter; batch or queue requests instead of sending parallel bursts.
Content or tool call rejected
Review sensitive content, malformed tool-call arguments, and JSON schema mismatches before retrying.
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.
01
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.
Model ID
02
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.
Protocols
03
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.
Migration
04
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.
Tool state
Prompt caching for repeated long context
Automatic cache reads use a dedicated lower rate when stable repository instructions, reference documents, and tool schemas can be reused.
1M-token context with a 65K max output limit
Keep connected code, specifications, documents, and agent state in one working context. Treat the limit as capacity rather than a target: retrieve relevant evidence and set task-appropriate output budgets.
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.
First-pass successAccepted deliverable rateRetry countOutput tokensCache-hit rateValid tool callsTime to accepted resultHuman correction and fallback rate
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
EvoLink
Measure task accuracy and accepted-result cost against Gemini 3.8 Flash’s higher token consumption. Then compare price, context, caching, and workload fit.
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 3.5-class route to compare with 3.8 Flash for classification, extraction, and high-throughput subagents.
Step up to the Pro tier when a task needs deeper reasoning than the Flash line can deliver.
Gemini 3.8 Flash
Input / output$0.675 / $3.375
Context1.05M
CachingAutomatic cache reads
Best forCoding, 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.
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 discounts are measured against Google’s standard list price of $1.500 input and $7.500 output per million tokens, not the introductory rate.
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 $0.750 input and $3.750 output per million tokens as an introductory price through December 31, 2026, with standard rates of $1.500 and $7.500 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.
