Gemini 3.5 Flash (High) API Access and Token Pricing
Use the exact model id antigravity/gemini-3-flash-agent on the antigravity/ route with a prepaid DoGPT key. The prices below come from the same catalog used by the public model directory.
Gemini 3.5 Flash (High) costs $1.50 per million input tokens and $9.00 per million output tokens through DoGPT, at the provider's published rate, with no margin added. A $10 prepaid balance covers roughly 6.7 million input tokens or 1.1 million output tokens, so a typical mixed workload lands somewhere between the two. The model answers to the exact id antigravity/gemini-3-flash-agent on the antigravity/ route; the prefix is part of the id. It accepts a context window of up to 1,048,576 tokens, returning at most 65,536 tokens per response. It is published as a single setting rather than a family of reasoning-effort tiers. Credit is prepaid and becomes the spending ceiling on your API key, so requests stop when it runs out rather than accruing a bill. Calls go straight from your code to https://api.dogpt.ir/v1; this storefront issues the key and holds the balance, and never sits in the request path.
How do I call Gemini 3.5 Flash (High) through DoGPT?
The branded endpoint accepts OpenAI-compatible chat completion requests. Set the environment variable to the reveal-once key from your account, then send the provider-qualified id exactly as shown. Inference travels directly to the API host. The storefront Worker does not proxy the request.
curl https://api.dogpt.ir/v1/chat/completions \
-H "Authorization: Bearer $DEVGPT_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"antigravity/gemini-3-flash-agent","messages":[{"role":"user","content":"Hello"}]}'A model-not-found response usually means the route prefix was removed or the id was copied from a vendor page instead of this catalog. Check the live directory before changing SDK settings.
How far does $10 go on Gemini 3.5 Flash (High)?
Ranked 13 of 51 by input price across the published catalog, with 12 models cheaper per input token and 38 more expensive.
$10 covers 6.667 million input tokens at this rate. $10 covers 1.111 million output tokens at this rate.
Real calls often contain both directions, so these are separate limits rather than a promise about the number of complete conversations. Tool calls, cached tokens, reasoning tokens, and long responses can change the mix reported by the upstream provider.
DoGPT adds no margin to the published rate and requires no subscription or monthly minimum. The wallet balance becomes the spending headroom on the upstream key. A request already in flight can finish slightly beyond the ceiling; the storefront absorbs that small overshoot instead of showing a negative customer balance.
What is the context window, and what can it do?
The catalog reports a 1,048,576-token context window. The largest published output is 65,536 tokens. These figures are copied from the model metadata feed and are not rounded guesses.
Capability badges report only flags the upstream catalog explicitly sets to true. An absent badge does not prove a feature is unsupported; it means the catalog did not make that claim. Test the request format required by your client before relying on an optional feature.
How do I use this model in Claude Code, Codex or Continue?
Tools that accept an OpenAI-compatible endpoint use https://api.dogpt.ir/v1. Anthropic-format tools use https://api.dogpt.ir because they append their own Messages API path. Each guide keeps that distinction in its copyable configuration.
What else is on the antigravity/ route?
- Claude Opus 4.6 (Thinking)$5.00 input / $25.00 output
- Claude Sonnet 4.6 (Thinking)$3.00 input / $15.00 output
- Gemini 2.5 Flash$0.30 input / $2.50 output
- Gemini 2.5 Flash Lite$0.10 input / $0.40 output
Prices and availability can change when the upstream catalog refreshes. The model directory and this page read the same cached snapshot, refreshed twice each hour. These figures were last read from upstream on 2026-08-18.