GPT-5.5 Pro
GPT-5.5 Pro is GPT-5.5 with more compute per answer: according to OpenAI, the same underlying model, using parallel compute at answer time. It came out on Apr 23, 2026 in ChatGPT and the next day on the API, at US$ 30 per million input tokens and US$ 180 per million output, with no cache discount. It costs 3 times the input and 3.6 times the output price of GPT-6 Astra, and on GPT-5.6 and GPT-6 models Pro became a mode billed at the model's normal rates. It has no API shutdown date; it makes sense for those who already rely on it.
GPT-5.5 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-5.5-pro(snapshotgpt-5.5-pro-2026-04-23)- Release
- Apr 23, 2026 in ChatGPT (Pro, Business and Enterprise plans); Apr 24, 2026 on the API
- Input
- US$ 30 per million tokens
- Output
- US$ 180 per million tokens
- Cache
- no cached input discount
- Above 272 thousand input tokens
- US$ 60 and US$ 270, per the pricing table
- Batch and Flex
- US$ 15 input and US$ 90 output
- Fast mode
- not offered
- Context window
- 1,050,000 tokens
- Max output
- 128 thousand tokens
- Knowledge cutoff
- Dec 1, 2025
- Reasoning
- effort
medium,high(default) andxhigh; nononeorlow - Where it runs
- OpenAI Responses API only, including batch; some requests take minutes
- Starting limit (Tier 1)
- 50 requests and 50,000 tokens per minute
- Status
- no deprecation notice on the API; in ChatGPT, the Pro options described today are GPT-5.6 Sol Pro and GPT-6 Pro
Where GPT-5.5 Pro does well, where it falls behind and who it is for
Where it does well
- Web research: 90.1% on BrowseComp, the top score in the launch table, against 84.4% for GPT-5.5, 89.3% for GPT-5.4 Pro and 85.9% for Gemini 3.1 Pro.
- Research math: 39.6% on FrontierMath Tier 4, against 35.4% for GPT-5.5, 38.0% for GPT-5.4 Pro and 22.9% for Claude Opus 4.7.
- Scientific data analysis: 33.2% on GeneBench, a genetics and quantitative biology test, against 25.0% for GPT-5.5 and 25.6% for GPT-5.4 Pro.
- Better than its predecessor at the same price: it costs the same as GPT-5.4 Pro and scores higher on 7 of the 8 tests where both have a figure in OpenAI's table.
Where it falls behind
- Price: US$ 30 and US$ 180 per million tokens, six times GPT-5.5, with no cache discount. Against GPT-6.1 Sol, 15 times the input and 18 times the output.
- Does not always beat GPT-5.5: on GDPval, an office work test, it scored 82.3% against 84.9% for regular GPT-5.5.
- Fewer tools: on the API it does not list computer use, tool search,
apply_patchor skills, which GPT-5.5 has. It also does not work on Chat Completions. - Slow, with low limits: OpenAI warns that some requests take several minutes and recommends background mode. At Tier 1, it allows 50 requests per minute, against 500 for GPT-5.5.
Who it makes sense for
- Hard, one-off questions where a mistake is expensive, such as a legal opinion or financial analysis, and the answer can wait minutes.
- Web research for hard-to-find information, where it posted the top launch score on BrowseComp.
- Integrations already built and validated on
gpt-5.5-pro, while you measure GPT-6 pro mode on the same tasks. - Data residency in the United Arab Emirates: OpenAI lists
gpt-5.5-pro-2026-04-23on the Responses API in that region, which needs extra approval, and no GPT-6 model.
When to pick another model
- Agents that operate a computer or edit code in a loop: the computer use and patch tools are missing, and each step can take minutes.
- Live chat, customer support and high volume: minutes of latency and a limit of 50 requests per minute at Tier 1.
- Long instructions repeated on every request: with no cache discount, every call pays the full US$ 30 per million input.
- New projects: on GPT-6 models, pro mode bills tokens at US$ 10 and US$ 50 on Astra or US$ 2 and US$ 10 on 6.1 Sol, although it uses more tokens than standard mode.
What "Pro" means in GPT-5.5 Pro
OpenAI describes GPT-5.5 Pro as the version of GPT-5.5 that uses more compute to think harder and give consistently better answers. The safety report goes further: it is the same underlying model, in a setting that uses parallel compute at answer time. That is why OpenAI uses GPT-5.5 safety results as a proxy for Pro, except where it judges the setting could change the risk.
In practice, Pro trades time and money for accuracy: it costs six times GPT-5.5 per token, has no cache discount, runs only on the Responses API (with batch) and defaults to high effort, against medium on GPT-5.5. Some requests take several minutes, and OpenAI suggests background mode.
In ChatGPT, Pro reached the Pro, Business and Enterprise plans on Apr 23. Today the help center describes GPT-5.6 Sol Pro and GPT-6 Pro, the latter powered by Astra, as the Pro options.
What OpenAI published at launch
The GPT-5.5 announcement table lists Pro on eight tests, all on knowledge, research and professional work. In the comparison below it leads five and loses three: GDPval to regular GPT-5.5, Humanity's Last Exam without tools to Claude Opus 4.7 and, with tools, to GPT-5.4 Pro.
| GPT-5.5 Pro OpenAI | GPT-5.5 OpenAI | GPT-5.4 Pro OpenAI | Claude Opus 4.7 Anthropic | |
|---|---|---|---|---|
| GDPval office work |
82.3% | 84.9% | 82.0% | 80.3% |
| Investment banking modeling spreadsheets, internal test |
88.6% | 88.5% | 83.6% | not measured |
| BrowseComp web research |
90.1% | 84.4% | 89.3% | 79.3% |
| GeneBench genetics data analysis |
33.2% | 25.0% | 25.6% | not measured |
| FrontierMath Tier 1 to 3 advanced math |
52.4% | 51.7% | 50.0% | 43.8% |
| FrontierMath Tier 4 research math |
39.6% | 35.4% | 38.0% | 22.9% |
| Humanity's Last Exam hard knowledge, no tools |
43.1% | 41.4% | 42.7% | 46.9% |
| Humanity's Last Exam hard knowledge, with tools |
57.2% | 52.2% | 58.7% | 54.7% |
Figures from the GPT-5.5 announcement (Apr 23, 2026), at xhigh effort. The dot marks the best in each row. "Not measured" means OpenAI published no figure for that model. The announcement has no Pro figures for coding, computer use or long context.
Is GPT-5.5 Pro still worth using in 2026?
On the API, GPT-5.5 Pro is still in the model index, with no entry on the deprecations page. In ChatGPT, GPT-5.5 is set to leave on Oct 14, 2026, but OpenAI's notice does not mention Pro separately.
The bigger change is in format. Since GPT-5.6, Pro is no longer a model with its own price but a mode, reasoning.mode: "pro", on the Responses API. OpenAI bills that mode's tokens at the chosen model's normal rates and warns that it uses more tokens than standard mode. Per token, GPT-6 Astra in pro mode costs 3 times less on input and 3.6 times less on output than GPT-5.5 Pro. What that comes to per task, OpenAI has not published.
If you already use GPT-5.5 Pro, there is no need to rush: OpenAI says existing Pro model IDs keep their behavior and pricing. If you are starting now, test pro mode on GPT-6.1 Sol or Astra first.
| GPT-5.5 Pro own model | GPT-5.4 Pro own model | GPT-6 Astra pro mode | GPT-6.1 Sol pro mode | |
|---|---|---|---|---|
| Input | US$ 30 | US$ 30 | US$ 10 | US$ 2 |
| Output | US$ 180 | US$ 180 | US$ 50 | US$ 10 |
| Cache read | no discount | no discount | US$ 1 | US$ 0.10 |
| Above 272 thousand tokens input and output |
US$ 60 and US$ 270 | US$ 60 and US$ 270 | US$ 20 and US$ 75 | US$ 4 and US$ 15 |
| Knowledge cutoff | Dec 1, 2025 | Aug 31, 2025 | Apr 30, 2026 | Apr 30, 2026 |
| How to call it on the API | model gpt-5.5-pro | model gpt-5.4-pro | reasoning.mode pro | reasoning.mode pro |
OpenAI pricing page and reasoning guide, checked on Oct 1, 2026. In pro mode the price per token is the model's normal rate, but each answer uses more tokens; OpenAI does not publish how many more.
What changes when moving to GPT-6 pro mode
Switching is optional, since the old identifier keeps working. For those who migrate, OpenAI's documentation raises these points:
- Pro is no longer a separate model: you call
gpt-6-astraorgpt-6.1-soland setreasoning.modeto"pro"on the Responses API. - Effort and mode are independent. On 6.1 Sol the default effort is
medium, while GPT-5.5 Pro defaulted tohigh; set the effort explicitly to compare at the same level. - Caching gives a discount again: reads at 10% of input (5% on 6.1 Sol) and writes at 1.25 times.
- GPT-6 Astra and GPT-6.1 Sol share Astra's safeguards stack, which includes misalignment monitoring. It can stop a task on the API, and there is no general way to resume the stopped conversation.
Examples of using GPT-5.5 Pro
The contract, the history and the question go in at once, and the model returns a structured analysis after a few minutes. It is the kind of work where, according to OpenAI, testers preferred GPT-5.5 Pro over GPT-5.4 Pro, especially in business and legal.
An immunologist at the Jackson Laboratory used GPT-5.5 Pro on a gene-expression dataset with 62 samples and nearly 28,000 genes and got a detailed research report. He said his team would have needed months to do the same. The account comes from OpenAI's announcement.
An analyst needs a figure scattered across obscure sources. BrowseComp measures this kind of persistent search, and GPT-5.5 Pro scored 90.1%, the top launch score.
Where GPT-5.5 Pro stands out and where it falls behind, by the numbers
GPT-5.5 Pro has published results on 13 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Jul 27, 2025 and Jul 22, 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 13 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| GPQA Diamond | 93.9% | #5 / 71 | #13 / 173 | 96.4% Gemini 4 Argon |
| AIME 2024 and 2025 (OTIS) | 100.0% | #1 / 71 (tied with 3) | #1 / 168 | 100.0% GPT-5.5 |
| Dtbench | 96.0% | #4 / 67 (tied with 2) | #12 / 148 | 98.9% Claude Opus 5.5 |
| Lmca | 53.94 | #7 / 65 | #13 / 121 | 68.23 Claude Opus 5.5 |
| CritPt | 30.6% | #2 / 60 | #8 / 112 | 32.3% GPT-5.6 Sol |
| Frontiermath Tiers 1 3 V2 | 87.7% | #2 / 50 | #8 / 86 | 93.7% GPT-6 Astra |
| ARC-AGI 2 | 84.6% | #4 / 42 | #12 / 83 | 95.0% GPT-6 Astra |
| ARC-AGI | 96.5% | #4 / 42 (tied with 1) | #10 / 81 | 98.5% Claude Fable 5 |
| Chess puzzles | 62.0% | #3 / 47 | #7 / 80 | 76.0% GPT-6 Astra |
| SimpleBench | 76.9% | #3 / 41 | #4 / 74 | 81.9% Claude Fable 5 |
| Frontiermath Tier 4 V2 | 78.0% | #3 / 40 | #9 / 67 | 100.0% GPT-6.1 Sol |
| FrontierMath | 51.0% | #2 / 37 | #2 / 65 | 51.7% GPT-5.5 |
| FrontierMath tier 4 | 39.6% | #1 / 37 | #1 / 48 | 39.6% GPT-5.5 Pro |
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-5.5 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 |
|---|---|---|---|---|---|
| Gemini 4 Argon | Google DeepMind | 164.56 | 1,608 | n/a | Sep 30, 2026 |
| Claude Opus 5 | Anthropic | 162.60 | 1,573 | 80.1 | Jul 24, 2026 |
| Claude Fable 5 | Anthropic | 162.13 | 1,542 | 83.0 | Jun 9, 2026 |
| GPT-5.5 Pro | OpenAI | 161.97 | n/a | n/a | Apr 23, 2026 |
| Kimi K3 | Moonshot | 157.42 | 1,528 | 79.2 | Jul 16, 2026 |
| Gemini 3.7 Flash | Google DeepMind | 157.22 | 1,551 | 78.8 | Aug 13, 2026 |
| Qwen 3.8 Max | Alibaba | 156.88 | 1,536 | 78.5 | Aug 2, 2026 |
GPT-5.5 Pro or Claude Fable 5?
Claude Fable 5 is the model from another lab closest to GPT-5.5 Pro in capability. Across the 11 tests both took, this is the score:
| Test | GPT-5.5 Pro | Claude Fable 5 |
|---|---|---|
| Chess puzzles | 62.0% | 47.0% |
| Frontiermath Tier 4 V2 | 78.0% | 90.2% |
| Lmca | 53.94 | 61.15 |
| GPQA Diamond | 93.9% | 85.9% |
| CritPt | 30.6% | 28.6% |
| SimpleBench | 76.9% | 81.9% |
| ARC-AGI 2 | 84.6% | 89.2% |
| Dtbench | 96.0% | 98.4% |
| ARC-AGI | 96.5% | 98.5% |
| Frontiermath Tiers 1 3 V2 | 87.7% | 87.0% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
What GPT-5.5 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-5.6 Sol | Jul 9, 2026 | 161.73 | 1,527 | 81.1 | API access |
| GPT-5.6 Terra | Jul 9, 2026 | 159.43 | 1,515 | 77.9 | API access |
| GPT-5.5 Instant | May 5, 2026 | 142.45 | 1,456 | n/a | API access |
| GPT-5.5 Pro | Apr 23, 2026 | 161.97 | n/a | n/a | API access |
| GPT-5.5 | Apr 23, 2026 | 158.95 | 1,533 | 80.2 | API 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 |
Have you used GPT-5.5 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-5.5 Pro
How much does GPT-5.5 Pro cost?
US$ 30 per million input tokens and US$ 180 per million output, with no cache discount. Above 272 thousand input tokens, the pricing table charges US$ 60 and US$ 270. In Batch or Flex, US$ 15 and US$ 90. There is no Fast mode.
What does GPT-5.5 Pro do differently from GPT-5.5?
According to OpenAI's safety report, it is the same underlying model using parallel compute at answer time. It thinks more, takes longer and costs six times more per token. At launch it beat GPT-5.5 on web research and math, but lost on GDPval.
Should I switch from GPT-5.5 Pro to GPT-6?
Per token, yes: in pro mode GPT-6 Astra charges US$ 10 and US$ 50 and GPT-6.1 Sol charges US$ 2 and US$ 10, against US$ 30 and US$ 180. But pro mode uses more tokens than standard mode, and OpenAI does not publish how many. Run the same tasks on both before deciding.
Where can I use GPT-5.5 Pro?
On the OpenAI API, only through the Responses API, with batch support, including in the United Arab Emirates region. In ChatGPT it reached the Pro, Business and Enterprise plans on Apr 23, 2026.
Is GPT-5.5 Pro being shut down?
On the API there is no published date and no entry on the deprecations page. In ChatGPT, OpenAI announced that GPT-5.5 leaves on Oct 14, 2026, without mentioning Pro separately.
Where does GPT-5.5 Pro rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, GPT-5.5 Pro ranks #10 out of 267 models, with an ECI of 161.97. 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-5.5 Pro better than Claude Fable 5?
Across the 11 tests both models took and Epoch AI compiled, GPT-5.5 Pro comes out ahead on 4, behind on 6 and tied on 1. 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-5.5 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-5.5 Pro are: FrontierMath tier 4 (#1 of 37); CritPt (#2 of 60); Frontiermath Tiers 1 3 V2 (#2 of 50). 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-5.5 announcement (includes GPT-5.5 Pro)
- OpenAI, GPT-5.5 Pro model page (documentation)
- OpenAI, API pricing
- OpenAI, GPT-5.5 system card (what Pro is)
- OpenAI, reasoning guide (pro mode)
- OpenAI, prompt caching (model differences)
- OpenAI, GPT-5.6 and GPT-6 Pro in ChatGPT (help center)
- OpenAI, Codex models and GPT-5.5 retirement
- OpenAI, API deprecations
- OpenAI, data residency (regions and models)
- OpenAI, misalignment monitoring (documentation)
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