# GPT-5.4 (`gpt-5.4`) Call gpt-5.4 on Lazu for $2.50 per million input tokens and $15.00 per million output tokens, the same as the OpenAI official price. 1.05M context, up to 128k output tokens, with Tool calling, Vision, Reasoning and Structured output. GPT-5.4 is OpenAI’s latest frontier model, unifying the Codex and GPT lines into a single system. It features a 1M+ token context window (922K input, 128K output) with support for... ## Facts - **Context:** 1.05M - **Max output:** 128k - **Knowledge cutoff:** 2025-08 - **Input:** Text · Image · PDF - **Output:** Text - **Endpoints:** /v1/chat/completions · /v1/responses ## Prices (USD per million tokens) | Lane | Input | Output | Cache read | Cache write 5m | 7-day availability | |---|---|---|---|---|---| | OpenAI official price | $2.50 | $15.00 | — | — | — | | OpenAI official, input over 272k tokens | $5.00 | $22.50 | — | — | — | | Stable | $2.50 | $15.00 | $0.25 | — | 97.5% | A request whose input tokens (cache reads and writes included) pass the threshold is billed entirely at the long-context prices, not only for the tokens above it. Prices checked 2026-10-03 17:20 UTC ## Capabilities Tool calling, Vision, Reasoning, Structured output, PDF input, Web search Reasoning effort: none, low, medium, high, xhigh ## Example request (POST /v1/chat/completions) ```bash curl -X POST https://api.lazu.ai/v1/chat/completions \ -H "Authorization: Bearer $LAZU_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-5.4", "messages": [{ "role": "user", "content": "Hello from Lazu" }] }' ``` Any OpenAI SDK works: set `base_url` to `https://api.lazu.ai/v1`. Docs: https://lazu.ai/docs ## Questions ### What is the difference between the stable and discount lanes? The stable lane runs on each maker's official API, is billed at the list price, has the highest availability and suits every use. The discount lane is a reverse-engineered route: a third-party supplier serves the model through its official client, not the official API. It costs far less, but availability varies, a failed request does not fall back to the stable lane, and the supplier may add its own system prompt. So use it only inside the matching agent tool: Claude models in Claude Code, GPT models in Codex, Gemini models in Antigravity. For your own code, or anywhere the output must be controlled exactly, use the stable lane. Both lanes work with the same key, and each key picks the lane per model. ### Can I call gpt-5.4 with the OpenAI SDK? Yes. Point the SDK's base URL at https://api.lazu.ai/v1 and use gpt-5.4 as the model name. ### How is it billed? Per token actually used — input, output and cache — deducted from your wallet balance. There is no monthly fee, and every request shows up itemized in your usage log.