GPT-5.4 Mini — Specs, Pricing & Limits
Removed from Codex for ChatGPT sign-in on 2026-08-31. API and API-key Codex use are unaffected.
Specs
| Status | Priced |
|---|---|
| Context window | 400,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | Aug 31, 2025 |
| Reasoning effort levels | none (default), low, medium, high, xhigh |
| Long-context rule | Not published |
| Cache writes | No additional charge — a written token bills as ordinary input |
| Codex with ChatGPT sign-in | Retired 2026-08-31 — replacement: GPT-6 Luna where available (Enterprise and Edu admins must enable Luna first). |
Pricing per 1M tokens, every published service tier
| Standard — short context | in $0.75 · cached $0.075 · writes — · out $4.50 |
|---|---|
| Batch — short context | in $0.375 · cached $0.0375 · writes — · out $2.25 |
| Flex — short context | in $0.375 · cached $0.0375 · writes — · out $2.25 |
| Fast — short context | in $1.50 · cached $0.15 · writes — · out $9.00 |
Rate limits by usage tier
| Free | Not supported |
|---|---|
| Build | 5,000 RPM · 2,000,000 TPM · 20,000,000 batch queue |
| Launch | 10,000 RPM · 10,000,000 TPM · 1,000,000,000 batch queue |
| Grow | 30,000 RPM · 180,000,000 TPM · 15,000,000,000 batch queue |
GPT-5.4 Mini is the mid-sized member of the GPT-5.4 generation, with a genuinely smaller context window rather than the full window at a discount — and since August 31, 2026 it is no longer available in Codex when you sign in with ChatGPT. OpenAI's named replacement there is GPT-6 Luna; on the API, and in Codex with your own key, GPT-5.4 Mini is unaffected.
Its smaller window sits above the long-context threshold that governs the larger models, yet no repricing rule is published for it and the pricing page carries no long-context band for it — checked directly rather than assumed. For a workload that fits inside the window, that is one less pricing mechanic to plan around.
It shares its rate-limit table with the Luna models, not with the rest of the GPT-5.4 family — a reminder that rate limits don't track model generation the way pricing does. That also makes the replacement comparison unusually clean: GPT-6 Luna runs on the same limits, at a lower rate on every published price cell, with a larger window and a newer cutoff.
Fast mode is published for this model at a premium over its Standard rate. It used to be one of the cheaper routes to lower latency on this roster; GPT-6 Luna's Fast tier now costs less than GPT-5.4 Mini's Standard tier.
Reasoning effort runs from none to xhigh, with none as the default — so an unspecified request does no reasoning.
The table above is computed directly from this site's facts module: pricing across every published service tier and the rate-limit table.
Verified 2026-10-28 against https://developers.openai.com/api/docs/models/gpt-5.4-mini.
Could not confirm: No >272K long-context repricing rule is stated on this model's own page, and the pricing page publishes no long-context band for it.
Checked: https://developers.openai.com/api/docs/models/gpt-5.4-mini · https://developers.openai.com/api/docs/pricing · https://learn.chatgpt.com/docs/models