What is GPT-5.2?
GPT-5.2 was OpenAI's flagship before GPT-5.4, and OpenAI now lists it as its previous flagship. It reads text and images, writes text, and works across a 400K-token window with up to 128K tokens of output; its knowledge stops at August 2025. Reasoning effort runs from none, the default, up to xhigh, a level that arrived with this model along with compaction for long conversations.
It is still on the API through Chat Completions, Responses and Batch, and it is not on OpenAI's removal list. OpenAI recommends GPT-6 Astra for new work, calls GPT-5.4 a drop-in replacement, and no longer offers GPT-5.2 in Codex for ChatGPT sign-ins.
Is GPT-5.2 still available?
Yes, on the OpenAI API, but OpenAI lists it as its previous flagship and points new work to newer models.
- API
- Still offered through Chat Completions, Responses and Batch, and not on OpenAI's list of models being removed.
- Codex
- Deprecated for people who sign in with ChatGPT; Codex with an API key can still use it.
- Variants
- The ChatGPT-tuned gpt-5.2-chat-latest and GPT-5.2-Codex have been removed from the API.
- Replacement
- OpenAI calls GPT-5.4 a drop-in replacement and recommends GPT-6 Astra as its current top model.
Who is GPT-5.2 best for?
GPT-5.2 is best for apps already built and tested on it, input-heavy jobs under 400K tokens, careful review and analysis, and batch work that can wait.
Apps already built on GPT-5.2
It is still served on Chat Completions, Responses and Batch, so tested prompts can keep running while you evaluate newer models.
Input-heavy jobs under 400K tokens
Its $1.75 input price is below GPT-5.4's $2.50 and GPT-6 Sol's $2, which counts when prompts are long and answers short.
Careful review and analysis
Developers use it to hunt bugs in code other models wrote, and analysts liked that it says when the evidence runs out.
Large jobs that can wait
Batch and Flex halve the price to $0.875 / $7 per million tokens for work that does not need an instant answer.
What's included with GPT-5.2?
Five reasoning levels up to xhigh, compaction for long conversations, OpenAI's built-in tools except computer use, half-price Batch and Flex, Fast mode and a Pro version.
- Five reasoning levels
- Effort runs from none, the default, through low, medium and high to xhigh, which first arrived with GPT-5.2.
- Compaction
- A compact endpoint in the Responses API shrinks long conversations so each turn sends less context.
- Built-in tools
- Web search, file search, code interpreter, image generation, hosted shell, apply patch, skills and MCP work through the Responses API; computer use does not.
- Batch and Flex
- Half-price processing: Batch returns results within 24 hours, and Flex answers more slowly and may turn requests away without charge.
- Fast mode
- Twice the standard rate, $3.50 / $28 per million tokens, for faster and steadier responses in user-facing apps.
- GPT-5.2 Pro
- A version that spends more compute on each answer, at $21 / $168, on the Responses API only.
How much does GPT-5.2 cost?
GPT-5.2 costs $1.75 per million input tokens and $14 per million output tokens on the OpenAI API, half that through Batch or Flex, and GPT-5.2 Pro costs $21 / $168.
- Cached input $0.0875
- Batch returns results within 24 hours
- Flex answers slower and may turn requests away, unbilled
- Cached input $0.175
- 400K-token context window, 128K output
- Reasoning effort from none to xhigh
- Cached input $0.35
- Faster, steadier responses for user-facing apps
- More compute per answer; some requests take minutes
- Batch: $10.50 / $84
- Responses API only
GPT-5.2 Pro has no cached-input price and is not offered on Flex or Fast mode. Tools such as web search are billed on top at their own rates. Full pricing on developers.openai.com. We checked these prices on .
How do readers rate GPT-5.2?
No one has rated GPT-5.2 yet. Be the first.
Sign in to rate. One rating per account.
GPT-5.2 pros and cons
Users credit it with careful bug-finding and honest answers, but it is slow at high effort, and newer OpenAI models give more for similar money.
Pros
- Several developers say it is good at finding bugs and fixing problems other models could not, which suits code review and hard debugging.
- Users doing analysis and clinical notes liked that it stops at the evidence and says what it does not know instead of filling gaps.
- Some developers found it needed less steering than Claude to finish coding tasks.
- Its $1.75 input price is lower than GPT-5.4's $2.50, and Batch or Flex halves every rate.
Cons
- Slow at higher reasoning settings: users report waits of several minutes, and some timeouts, on prompts that Gemini or Grok answered faster.
- GPT-5.2 Pro costs 12 times the base rate at $21 / $168, and early users questioned whether its gains justify that.
- Some developers still prefer Claude for front-end and UI code.
- Smaller and older than OpenAI's current models: a 400K window, no computer use or tool search, and knowledge that stops at August 2025.
What are the best GPT-5.2 alternatives?
Five tools on Recaply do GPT-5.2's job. DeepSeek-V3.2 and MiMo-V2-Flash are open source, so you can run them on your own server.
GPT-5.4PaidChoose it when you need more than 400K tokens or computer use; OpenAI calls it a drop-in replacement for GPT-5.2, with a 1.05M window at $2.50 / $15.
GPT-6 AstraPaidChoose it when quality matters more than cost; it is the model OpenAI recommends in place of GPT-5.2, with a 1.05M window and knowledge to April 2026, at $10 / $50.
Claude Sonnet 5FreeChoose it when you want a similar price outside OpenAI: $2 / $10 per million tokens, a 1M-token window at that same rate, and knowledge to January 2026.
DeepSeek-V3.2Open SourceChoose it when you want an MIT-licensed reasoning model for agents that keeps thinking during tool calls on your own servers.
MiMo-V2-FlashOpen SourceChoose it when you need an open model for coding and reasoning that you can host and fine-tune yourself.
Questions about GPT-5.2
What is GPT-5.2 Pro?
GPT-5.2 Pro is a version that spends more compute on each answer for hard problems. It costs $21 / $168 per million tokens, runs only on the Responses API, and some requests take several minutes. OpenAI now points to GPT-5.5 Pro instead.
Who is GPT-5.2 for?
GPT-5.2 is best for apps already built and tested on it, input-heavy jobs under 400K tokens, careful review and analysis, and batch work that can wait.
Is GPT-5.2 free?
No. GPT-5.2 is a paid tool, from $0.875 / $7 per million input / output tokens.
How much does GPT-5.2 cost?
GPT-5.2's paid plans are Batch or Flex at $0.875 / $7 per million input / output tokens, Standard at $1.75 / $14 per million input / output tokens, Fast mode at $3.50 / $28 per million input / output tokens and GPT-5.2 Pro at $21 / $168 per million input / output tokens.
Is there an open source alternative to GPT-5.2?
Yes. DeepSeek-V3.2 and MiMo-V2-Flash are open source tools that do the same job, so you can run your own.
What are the best GPT-5.2 alternatives?
The closest alternatives on Recaply are GPT-5.4, GPT-6 Astra, Claude Sonnet 5, DeepSeek-V3.2 and MiMo-V2-Flash.
Where can I try GPT-5.2?
On its official website, developers.openai.com.
GPT-5.2 vs GPT-5.4: which should I use?
GPT-5.4 for most new work. OpenAI calls it a drop-in replacement for GPT-5.2, and it adds a 1.05M-token window, computer use and tool search. GPT-5.2 is cheaper on input, $1.75 against $2.50 per million tokens, so it can still suit long prompts under 400K tokens.
Can I use GPT-5.2 in Codex?
Only with an API key. OpenAI has deprecated GPT-5.2 in Codex for people who sign in with ChatGPT. Codex signed in with an API key offers the models available to that key, and GPT-5.2 is still on the API, billed per token.
Why does GPT-5.2 reject temperature or top_p?
Temperature, top_p and logprobs only work when GPT-5.2's reasoning effort is set to none; with any other effort level the request returns an error. Use reasoning effort, verbosity and max_output_tokens to shape answers instead.
Our verdict
GPT-5.2 is worth keeping only where it already works; new projects should start on a newer OpenAI model.
Users rate it as careful, but it is slow at high effort, its window stops at 400K tokens, and OpenAI's GPT-6 Sol costs about the same at $2 / $10 with a 1.05M window and newer knowledge.
- Skip it for
- New builds, agents that need computer use, and anything that must answer quickly.
- Check first
- Before you move off it, run your own prompts on GPT-5.4, which OpenAI says works as a drop-in replacement.
SubQ
GPT-5.4 mini & nano
Grok 4.20
Nemotron 3
Nemotron 3 Super
NVIDIA Nemotron 3 Ultra