Cette section n'est pas encore traduite. Le contenu ci-dessous est affiché en anglais, tel qu'il a été publié. Lire la version anglaise.
OpenAI · Text model · GA
OpenAI: our strongest mini model yet for coding, computer use and subagents developers.openai.com (opens in a new tab)
Compare with similar models Open the Model Finder
Figures checked 1 Nov 2026. A dash means not published on the sources we checked, so it stays blank rather than estimated.
OpenAI names these settings reasoning.effort. Price per token is the same at every level. Higher effort uses more reasoning tokens, billed as output. developers.openai.com (opens in a new tab)
| Level | When to use it, per OpenAI | AA index | Tokens/s |
|---|---|---|---|
| none default | Latency-critical tasks that do not benefit from reasoning, such as voice, fast retrieval and classification developers.openai.com (opens in a new tab) | – | – |
| low | Efficient reasoning with a modest latency increase. Data analysis, drafting, execution-oriented coding and support chat developers.openai.com (opens in a new tab) | – | – |
| medium | When quality and reliability matter. The default for most workloads and a balanced point for latency, performance and cost developers.openai.com (opens in a new tab) | – | – |
| high | Hard reasoning, complex debugging, deep planning and high-value tasks where quality matters more than latency developers.openai.com (opens in a new tab) | – | – |
| xhigh | Deep research, asynchronous and long agentic runs. Only when your evals show a clear benefit for the extra latency and cost developers.openai.com (opens in a new tab) | – | – |
Our rule of thumb: Simple or speed-critical work: low. Standard work: the lab's default. Complex work: high. Use xhigh or max only when your own tests show a gain, as OpenAI and Anthropic both advise. developers.openai.com (opens in a new tab)platform.claude.com (opens in a new tab)
US$ per 1M tokens, standard tier, short prompts. Output left blank where not published.
| Model | Input US$/1M | Output US$/1M |
|---|---|---|
| GPT-5.4 mini | 0.75 | 4.5 |
| Grok 4.20 | 1.25 | 2.5 |
| Grok 4.20 Multi-Agent (beta) | 1.25 | 2.5 |
| Grok 4.3 | 1.25 | 2.5 |
| Gemini 3.6 Flash | 0.75 | 3.75 |
OpenAI: Remote MCP servers as a tool in the Responses API developers.openai.com (opens in a new tab). These official remote MCP servers fit that rule. We have not tested each server with this model. A server that publishes a client list is left out when this lab is not on it.
Same type, closest on quality index and blended price. A similarity, not a ranking.
Compare GPT-5.4 mini with Grok 4.20 and Grok 4.20 Multi-Agent (beta)
All 109 AI models · AI agents · How we check sources
OpenAI 
CNBC confirmed OpenAI is now seeking up to US$20bn, down from about US$30bn, and the FT put the post-money valuation at US$1.25tn. The round is still open. Less fresh equity means more scrutiny of every gigawatt OpenAI has signed for.
Source: CNBC (opens in a new tab)
OpenAI
OpenAI says Terra matches GPT-6 Astra on Agents' Last Exam at a tenth of the price per token and finishes tasks over 3x faster.
Source: OpenAI (opens in a new tab)
