GPT-5.2 Pro
GPT-5.2 Pro is the Pro version of GPT-5.2, built for hard questions where a better answer is worth the wait, released by OpenAI on Dec 11, 2025 in ChatGPT and on the API. It costs US$ 21 per million input tokens and US$ 168 per million output, 12 times GPT-5.2, and was the first model to cross 90% on ARC-AGI-1, according to OpenAI. It left ChatGPT on Jun 12, 2026. It is still on the API with no shutdown date, but the model page itself recommends GPT-5.5 Pro instead.
GPT-5.2 Pro in official numbers
Data published by the maker in its documentation and announcement. The source of each item is at the end of the page.
- Maker
- OpenAI
- API identifier
gpt-5.2-pro(snapshotgpt-5.2-pro-2025-12-11)- Release
- Dec 11, 2025, in ChatGPT (paid plans) and on the API
- Input
- US$ 21 per million tokens
- Output
- US$ 168 per million tokens
- Cache
- no cached input discount
- Batch
- US$ 10.50 and US$ 84 in the pricing table, although the model page says the Batch API is not supported
- Fast mode
- not offered
- Context window
- 400 thousand tokens
- Max output
- 128 thousand tokens
- Knowledge cutoff
- Aug 31, 2025
- Reasoning
- effort
medium,highandxhigh - Where it runs
- Responses API only; some requests take minutes, and OpenAI suggests background mode
- Tools on the API
- functions, file search, image generation, MCP and web search
- Starting limit (Tier 1)
- 500 requests and 30,000 tokens per minute
- Status
- out of ChatGPT since Jun 12, 2026; on the API, a "previous pro model" with no deprecation notice
Where GPT-5.2 Pro does well, where it falls behind and who it is for
Where it does well
- Abstract reasoning: 90.5% on ARC-AGI-1, the first model to cross 90%, according to OpenAI, which says it cut the cost of that score range by roughly 390 times since o3-preview (87%).
- Office work: 74.1% wins or ties on GDPval, against 70.9% for GPT-5.2 Thinking; on clear wins, 60.0% against 49.8%.
- Web research: 77.9% on BrowseComp, against 65.8% for GPT-5.2 Thinking.
- Competition math: 100% on AIME 2025 and on HMMT February 2025, with no tools.
- The cheapest Pro: US$ 21 and US$ 168, against US$ 30 and US$ 180 for GPT-5.4 Pro and GPT-5.5 Pro.
Where it falls behind
- Overtaken by its successor: GPT-5.4 Pro scored higher on all 9 tests where both appear in the March table, such as ARC-AGI-2 (83.3% against 54.2%).
- Smaller window: 400 thousand tokens, against 1.05 million for GPT-5.4 Pro and GPT-5.5 Pro.
- No cache discount: every call pays US$ 21 per million input tokens, even with a repeated instruction.
- Slow, with a low limit: some requests take minutes, and the Tier 1 limit is 30,000 tokens per minute.
- Few tools: it does not list code interpreter, hosted shell or computer use.
Who it makes sense for
- Older integrations pinned to the
gpt-5.2-pro-2025-12-11snapshot, while the switch is being tested. - Reproducing results published with GPT-5.2 Pro, such as analyses run between December 2025 and March 2026.
- Teams that need a dedicated Pro identifier and want the lowest price per token among OpenAI's Pro models, accepting lower scores.
- A staged migration: keep 5.2 Pro while you measure GPT-5.5 Pro and GPT-6 pro mode on the same questions, since no shutdown date has been published.
When to pick another model
- New projects: the model page recommends GPT-5.5 Pro.
- Documents above 400 thousand tokens, which do not fit the window.
- Live chat, customer support and volume: requests that take minutes and a low tokens-per-minute limit.
- Agents that need to run code, use a shell or operate screens: those tools are not on the list.
- ChatGPT: the model left on Jun 12, 2026, and old conversations moved to the matching GPT-5.5 model.
What "Pro" means in GPT-5.2 Pro
In ChatGPT, OpenAI presented GPT-5.2 Pro as its smartest and most trustworthy option for hard questions where a better answer is worth the wait, with fewer major errors in early testing. On the API it runs only on the Responses API, and the company warns that some requests take several minutes. OpenAI did not detail how Pro works internally.
The launch brought two changes for API users: the reasoning setting of Pro became adjustable, and both Pro and GPT-5.2 gained the xhigh effort, the fifth level. The price is 12 times GPT-5.2, with no cache discount. The GPT-5.2 safety report does not cover Pro separately; in it, GPT-5.2 Thinking is rated High in biology and chemistry and below High in cybersecurity.
Since GPT-5.6, Pro is also a mode of the newer models, turned on with reasoning.mode set to "pro" and billed at each model's normal rates. OpenAI says existing Pro model IDs keep their behavior and pricing.
What OpenAI published at launch
The GPT-5.2 announcement included a full table for GPT-5.2 Thinking and only a subset for Pro. On the tests where both appear, Pro is ahead on all of them except AIME 2025, where it ties. There are no Pro scores for coding, long context or vision.
| GPT-5.2 Pro OpenAI | GPT-5.2 Thinking OpenAI | GPT-5.1 Thinking OpenAI | |
|---|---|---|---|
| GDPval wins or ties |
74.1% | 70.9% | 38.8% |
| GDPval clear wins |
60.0% | 49.8% | 35.5% |
| Investment banking spreadsheets internal test |
71.7% | 68.4% | 59.1% |
| BrowseComp web research |
77.9% | 65.8% | 50.8% |
| GPQA Diamond PhD-level science |
93.2% | 92.4% | 88.1% |
| Humanity's Last Exam no tools |
36.6% | 34.5% | 25.7% |
| Humanity's Last Exam with search and Python |
50.0% | 45.5% | 42.7% |
| HMMT, February 2025 competition math |
100.0% | 99.4% | 96.3% |
| AIME 2025 competition math |
100.0% | 100.0% | 94.0% |
| ARC-AGI-1 abstract reasoning |
90.5% | 86.2% | 72.8% |
| ARC-AGI-2 abstract reasoning, harder version |
54.2% | 52.9% | 17.6% |
Appendix of the GPT-5.2 announcement (Dec 11, 2025), at each model's maximum API effort: xhigh for GPT-5.2 Pro and Thinking, high for GPT-5.1 Thinking. On ARC-AGI-2, GPT-5.2 Pro was measured at high. On GDPval, the GPT-5.1 Thinking column shows the GPT-5 score. The dot marks the best in each row; AIME is a tie.
Is GPT-5.2 Pro still worth using in 2026?
For a new project, no. The model page itself calls it the previous pro model and recommends GPT-5.5 Pro. Twelve weeks after launch, GPT-5.4 Pro arrived with higher scores on all nine tests where both appear and a 1.05 million token window, at US$ 30 and US$ 180.
In ChatGPT, the GPT-5.2 models left on Jun 12, 2026, Pro included. On the API, gpt-5.2-pro has no entry on the deprecations page, and generally available models get at least six months of notice.
Price per token is misleading. GPT-5.2 Pro costs 2.1 times GPT-6 Astra on input and 3.36 times on output, and 10.5 times and 16.8 times GPT-6.1 Sol. Pro mode on those models uses more tokens than standard mode, and OpenAI does not publish how many more, so only a test with your own questions shows the cost per task.
| GPT-5.2 Pro Dec 2025 | GPT-5.4 Pro Mar 2026 | GPT-5.5 Pro Apr 2026 | |
|---|---|---|---|
| GDPval office work |
74.1% | 82.0% | 82.3% |
| Investment banking modeling internal test |
71.7% | 83.6% | 88.6% |
| BrowseComp web research |
77.9% | 89.3% | 90.1% |
| FrontierMath Tier 4 research math |
31.3% | 38.0% | 39.6% |
| Humanity's Last Exam with tools |
50.0% | 58.7% | 57.2% |
| ARC-AGI-2 abstract reasoning |
54.2% | 83.3% | not measured |
| Price per million tokens input and output |
US$ 21 and US$ 168 | US$ 30 and US$ 180 | US$ 30 and US$ 180 |
| Context window | 400 thousand | 1.05 million | 1.05 million |
| Knowledge cutoff | Aug 31, 2025 | Aug 31, 2025 | Dec 1, 2025 |
GPT-5.2 Pro and GPT-5.4 Pro scores: GPT-5.4 announcement (Mar 5, 2026). GPT-5.5 Pro scores: GPT-5.5 announcement (Apr 23, 2026). At xhigh effort, except GPT-5.2 Pro on ARC-AGI-2 (high). Prices and windows from OpenAI's documentation, checked on Oct 1, 2026. The dot marks the best in each score row.
What changes when replacing GPT-5.2 Pro
Since older Pro model IDs keep their behavior and pricing, according to OpenAI, there is no rush. Anyone who decides to switch has two routes:
- To GPT-5.5 Pro: switch to
gpt-5.5-pro. Input and output rise to US$ 30 and US$ 180, the window goes to 1.05 million tokens, the default effort ishighand code interpreter and hosted shell come in. It has no cache discount either. - To GPT-6 pro mode: call
gpt-6-astraorgpt-6.1-solwithreasoning.modeset to"pro". Tokens cost US$ 10 and US$ 50 or US$ 2 and US$ 10, with a cache discount, but each answer uses more tokens. - On GPT-6 models, caching uses
prompt_cache_options.ttlset to"30m", and writes cost 1.25 times input. - Knowledge moves from August 2025 to December 2025 on GPT-5.5 Pro and to April 2026 on GPT-6 Astra and GPT-6.1 Sol.
Examples of using GPT-5.2 Pro
A graduate-level physics, chemistry or biology question, with no lookup. On GPQA Diamond, GPT-5.2 Pro scored 93.2%, against 92.4% for GPT-5.2 Thinking. GPT-5.4 Pro later reached 94.4% on the same test.
OpenAI says researchers used GPT-5.2 Pro on an open question in statistical learning theory, in a narrow, well-specified setting. The model proposed a proof that the authors then verified and outside experts reviewed.
An analyst asks the model to find hard-to-locate data across many sources. On BrowseComp, GPT-5.2 Pro scored 77.9%; GPT-5.5 Pro reaches 90.1%, at US$ 30 and US$ 180.
Where GPT-5.2 Pro stands out and where it falls behind, by the numbers
GPT-5.2 Pro has published results on 6 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Mar 16, 2025 and Mar 11, 2026 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
It is not in the bottom half of its generation on any test with 6 or more models compared.
See all 6 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| Frontiermath Tiers 1 3 V2 | 74.0% | #3 / 24 | #23 / 86 | 93.7% GPT-6 Astra |
| ARC-AGI 2 | 54.2% | #7 / 41 | #33 / 83 | 95.0% GPT-6 Astra |
| ARC-AGI | 90.5% | #5 / 41 | #25 / 81 | 98.5% Claude Fable 5 |
| Chess puzzles | 60.0% | #1 / 25 | #8 / 80 | 76.0% GPT-6 Astra |
| SimpleBench | 57.4% | #12 / 32 | #30 / 74 | 81.9% Claude Fable 5 |
| Frontiermath Tier 4 V2 | 46.0% | #3 / 17 | #23 / 67 | 100.0% GPT-6.1 Sol |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare GPT-5.2 Pro with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Muse Spark 1.3 | Meta AI | 156.76 | 1,528 | 81.6 | Sep 2, 2026 |
| Claude Sonnet 5 | Anthropic | 156.36 | 1,491 | 76.0 | Jun 30, 2026 |
| Claude Opus 4.7 | Anthropic | 156.32 | 1,532 | 76.5 | Apr 16, 2026 |
| Grok 4.6 | xAI | 156.31 | 1,475 | 78.0 | Aug 12, 2026 |
| GPT-5.2 Pro | OpenAI | 156.03 | n/a | n/a | Dec 11, 2025 |
| GLM-5.3 | Z.ai (Zhipu AI) | 155.91 | n/a | 76.1 | Aug 14, 2026 |
| DeepSeek V4 Pro 0813 | DeepSeek | 155.29 | n/a | 77.4 | Aug 13, 2026 |
GPT-5.2 Pro or Grok 4.6?
Grok 4.6 is the model from another lab closest to GPT-5.2 Pro in capability. Across the 6 tests both took, this is the score:
| Test | GPT-5.2 Pro | Grok 4.6 |
|---|---|---|
| Chess puzzles | 60.0% | 33.0% |
| Frontiermath Tier 4 V2 | 46.0% | 31.7% |
| SimpleBench | 57.4% | 75.9% |
| ARC-AGI 2 | 54.2% | 67.1% |
| Frontiermath Tiers 1 3 V2 | 74.0% | 66.0% |
| ARC-AGI | 90.5% | 87.5% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
What GPT-5.2 Pro costs to deliver each task
Price per token is what is on the price sheet; cost per task is what lands on the invoice, because it depends on how many tokens the model spends. Each point is an effort level, published by whoever ran the test and compiled by Epoch AI. The further up and to the left, the more score per dollar.
ARC-AGI 2
The score scale does not start at zero.
ARC-AGI
The score scale does not start at zero.
Other models from OpenAI
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| GPT-5.4 Mini | Mar 17, 2026 | 148.77 | 1,446 | 66.4 | API access |
| GPT-5.4 Nano | Mar 17, 2026 | 146.12 | 1,404 | 69.6 | API access |
| GPT-5.4 Pro | Mar 5, 2026 | 159.00 | n/a | n/a | API access |
| GPT-5.4 | Mar 5, 2026 | 156.72 | 1,519 | 78.0 | API access |
| GPT-5.3 Codex | Feb 5, 2026 | 156.76 | n/a | n/a | API access |
| GPT-5.2 Pro | Dec 11, 2025 | 156.03 | n/a | n/a | API access |
| GPT-5.2 | Dec 11, 2025 | 153.46 | 1,454 | 74.6 | API access |
| GPT-5.1 | Nov 13, 2025 | 149.76 | 1,495 | n/a | API access |
| GPT-5 Pro | Oct 7, 2025 | 150.74 | n/a | n/a | API access |
Have you used GPT-5.2 Pro? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about GPT-5.2 Pro
How much does GPT-5.2 Pro cost?
US$ 21 per million input tokens and US$ 168 per million output, with no cache discount. The pricing table lists batch at US$ 10.50 and US$ 84, but the model page says the Batch API is not supported. The price has not changed since launch.
Is GPT-5.2 Pro still available?
In ChatGPT, no: the GPT-5.2 models left on Jun 12, 2026. On the API, yes, through the Responses API, with no published shutdown date. The model page recommends GPT-5.5 Pro.
What is the difference between GPT-5.2 and GPT-5.2 Pro?
Pro is OpenAI's option for hard questions where a better answer is worth the wait. It costs 12 times more per token, runs only on the Responses API and takes longer. At launch it scored higher than GPT-5.2 Thinking on office work, web research, science and abstract reasoning, and tied on AIME 2025.
Is it worth switching GPT-5.2 Pro for GPT-5.5 Pro?
In OpenAI's figures, GPT-5.5 Pro scores higher on all six tests where both appear, has a 1.05 million token window and knowledge through December 2025. It costs US$ 30 and US$ 180, against US$ 21 and US$ 168, so the switch pays off if the extra quality covers the difference in your case.
GPT-5.2 Pro or GPT-6 pro mode?
Pro mode on GPT-6 Astra and GPT-6.1 Sol bills tokens at each model's normal rates, well below the US$ 21 and US$ 168 of GPT-5.2 Pro, but uses more tokens than standard mode, and OpenAI does not publish how many more. A test with your own questions decides. The independent scores are at the top of this page.
Where does GPT-5.2 Pro rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, GPT-5.2 Pro ranks #29 out of 267 models, with an ECI of 156.03. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is GPT-5.2 Pro better than Grok 4.6?
Across the 6 tests both models took and Epoch AI compiled, GPT-5.2 Pro comes out ahead on 4, behind on 2 and tied on 0. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is GPT-5.2 Pro best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of GPT-5.2 Pro are: Chess puzzles (#1 of 25); Frontiermath Tiers 1 3 V2 (#3 of 24); ARC-AGI (#5 of 41). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
- OpenAI, GPT-5.2 announcement (includes GPT-5.2 Pro)
- OpenAI, GPT-5.2 Pro model page (documentation)
- OpenAI, API pricing
- OpenAI, GPT-5.2 system card
- OpenAI, GPT-5.4 announcement (table with GPT-5.2 Pro and GPT-5.4 Pro)
- OpenAI, GPT-5.5 announcement (table with GPT-5.5 Pro)
- OpenAI, GPT-5.5 Pro model page (documentation)
- OpenAI, reasoning guide (pro mode)
- OpenAI, ChatGPT release notes (retirement on Jun 12, 2026)
- OpenAI, API deprecations
- OpenAI, prompt caching guide
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
← See the full ranking