Transparent pricing with no hidden fees. Pay as you go.
gemini-3.8-flashgemini-3.8-flashgemini-3.8-flash| Model | Spec | PoYo price | Official price | You save |
|---|---|---|---|---|
| gemini-3.8-flash | Input | $0.600/1M tokens 120 credits /1M tokens | $0.750/1M tokens Official | 20% |
Cached input | $0.060/1M tokens 12 credits /1M tokens | $0.075/1M tokens Official | 20% | |
Output (including thinking tokens) | $3.00/1M tokens 600 credits /1M tokens | $3.75/1M tokens Official | 20% |
* Actual fees are based on the final output.
Complete guide to using Gemini 3.8 Flash API for Autonomous Agents & Long-Horizon Coding
Gemini 3.8 Flash is Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. It supports a 1,048,576-token input context and up to 65,536 output tokens.
Use gemini-3.8-flash through /v1/chat/completions or the Gemini-native /v1beta/models path. PoYo charges 120 input, 12 cached-input, and 600 output credits per million tokens—20% below Google's introductory Standard rates.
Official Google demonstrations showcase Gemini 3.8 Flash building interactive 3D worlds, functional retro software, scientific visualizations, and hardware teardowns.
01
Using looping instructions in Google Antigravity, Gemini 3.8 Flash generates full 3D interactive levels with puzzles, procedural textures, and complete environmental gameplay.
02
Gemini 3.8 Flash builds a fully playable DOS version of Google Maps in a single prompt, complete with location search, turn-by-turn routing, and terminal-rendered Street View.
03
Synthesizes real datasets from the U.S. Geological Survey into real-time 3D topographic slices, 2D projections, and domain-grounded scientific explanations.
04
Generates realistic Three.js visualizers of physically-proportioned hardware teardowns, automatically decomposing devices into inspectable exploded layers with interactive sliders.
Resolve multi-file refactoring, autonomous debugging, and complex engineering tasks end-to-end with high diligence.
Deploy reliable agents capable of recursive reasoning, multi-step planning, tool orchestration, and self-verification.
Analyze complex financial models, legal documents, scientific reports, and enterprise data with high domain precision.
Generate full 3D interactive applications, procedural games, and retro software with code and visual grounding.
Process text, high-resolution images, video, audio, and technical PDFs with 1M-token context recall.
Select low, medium, or high thinking effort levels to flexibly balance execution speed, token cost, and reasoning depth.
September 2026 benchmarks compare Gemini 3.8 Flash against Gemini 3.7 Flash, Claude Sonnet 5, and GPT-5.6 Terra across software engineering, agentic autonomy, document intelligence, and expert reasoning.
| Benchmark | Gemini 3.8 Flash | Gemini 3.7 Flash | Claude Sonnet 5 | GPT-5.6 Terra | Notes |
|---|---|---|---|---|---|
| DeepSWE v1.1 | 73.7% | 65.3% | 53.8% | 69.6% | Long-horizon software engineering |
| Terminal-Bench 2.1 | 89.4% | 85.8% | 80.4% | 87.4% | Agentic terminal coding |
| HLE-Verified | 54.9% | 53.6% | 31.0% | 51.1% | Multidisciplinary expert reasoning |
| Vals Finance Agent v2 | 61.4% | 59.0% | 56.2% | 60.1% | Financial workflow autonomy |
| Harvey Legal Agent Benchmark | 10.0% | 8.8% | 7.9% | 9.2% | Full legal task resolution |
| WebDev Arena (Elo) | 1612 | 1588 | 1541 | 1523 | Interactive web development |
| Artificial Analysis Intelligence Index | 59 | 56 | 55 | 57 | Composite intelligence |
| AutomationBench | 34.2% | 30.4% | 10.7% | 23.6% | Enterprise workflow automation |
| GDP.pdf | 36.8% | 34.0% | 28.0% | 24.7% | Complex document processing |
| GDM-MRCR v2 (128k) | 98.2% | 97.0% | 81.5% | 93.5% | Long-context retrieval & recall |
Official evaluation results as of September 2, 2026. Benchmarks follow standard published evaluation harnesses.
1. Create an API key. Sign in and create a PoYo key.
2. Choose a format. Use POST /v1/chat/completions for OpenAI-compatible clients, or POST /v1beta/models/gemini-3.8-flash:generateContent for Gemini-native payloads. Streaming is available through :streamGenerateContent.
3. Set the model. Send gemini-3.8-flash as the model ID. Read the Gemini API documentation.
gemini-3.8-flash model ID with either supported endpoint format./v1/chat/completions for OpenAI-compatible requests or /v1beta/models/gemini-3.8-flash:generateContent for Gemini-native requests. Both non-streaming and streaming flows are supported.Use OpenAI-compatible Chat Completions or Gemini-native GenerateContent with one model ID.
Evaluate and ship Gemini alongside other models without managing another integration credential.
See input, cached-input, and output rates in both credits and US-dollar equivalents.
Test both request formats, review service status, and track usage before moving to production.