GPT-5.2
GPT-5.2 is OpenAI's best model for coding and agentic tasks across industries.
- Input price
- $1.575list $1.7510% off
- Output price
- $12.60list $14.0010% off
- Context
- 400K
- Max output
- 128K
- Capability
- 155 (#45 of 82)
- Human preference
- 1412
- 30-day uptime
- —
- Released
- 2025-12-11
- Knowledge cutoff
- 2025-08-31
Providers
This model is served by the dedicated providers below. We pick the provider for each request, so provider preferences don’t apply, including a provider named in your request. Prices in USD per 1M tokens. Need a custom setup? Contact [email protected].
- SuccessTTFT
- Azure AI FoundryGlobal——
- OpenAIUnited States——
Pricing
Prices for this model. Each request is billed at these prices, whichever provider serves it.
| Item | Standard | Standard Discount | Batch | Batch Discount | Priority | Priority Discount | Price | Price Discount | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Inputper 1M tokens | $1.575 | list $1.75 | 10% off | $0.7875 | list $0.875 | 10% off | $3.15 | list $3.50 | 10% off | — | ||
| Outputper 1M tokens | $12.60 | list $14.00 | 10% off | $6.30 | list $7.00 | 10% off | $25.20 | list $28.00 | 10% off | — | ||
| Cache readper 1M tokens | $0.1575 | list $0.175 | 10% off | $0.078752 | list $0.0875 | 10% off | $0.315 | list $0.35 | 10% off | — | ||
| Web searchper request | — | — | — | $0.0126 | list $0.014 | 10% off | ||||||
| Estimated cost of common use (500 output tokens each) | ||||||||||||
| One chat message3K input tokens | $0.011 | — | $0.0221 | — | ||||||||
| Agent first turn100K tokens of context, none cached | $0.164 | — | $0.328 | — | ||||||||
| Each later agent turncontext read from cache, plus 2K new tokens | $0.0252 | — | $0.0504 | — | ||||||||
| 20-turn agent taskfirst turn plus 19 more | $0.643 | — | $1.29 | — | ||||||||
Try it
Runs on your balance and is billed like any API request.
Similar models
Other models with the same output types at a similar price.
Human preference ratings are from the LMArena Text Arena (CC BY 4.0, as of 2026-11-18): Elo ratings from blind votes, on a different scale from capability scores. lmarena.ai/leaderboard/text