# GPT-5.5

GPT-5.5 is an AI model released by OpenAI on Apr 23 2026. It has a 1.05M context window. Tracked results include 73.1% on Expert-SWE (Internal), 84.9% on GDPval (win/tie rate) and 36.2% on Blueprint-Bench 2.

## Facts

| Field | Value |
| --- | --- |
| Model | GPT-5.5 |
| Developer | OpenAI |
| Release date | Thursday, Apr 23 2026 |
| Licensing | Proprietary |
| Context window | 1.05M |

## API pricing

All rates in USD per 1,000,000 tokens, pay-as-you-go.

| Tier | Input | Cached input | Output |
| --- | --- | --- | --- |
| Up to 272K input tokens | $5.00 | $0.50 | $30.00 |

Verified August 18, 2026 against the first-party source: https://developers.openai.com/api/docs/models/gpt-5.5

## Tracked benchmark scores

| Benchmark | Score | Source | What it measures |
| --- | --- | --- | --- |
| BullshitBench v2 | 47% | [BullshitBench](https://github.com/petergpt/bullshit-benchmark) | Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. |
| Gray Swan IPI (k = 1) | 3% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. Gray Swan's indirect prompt injection benchmark measures how often such an attack succeeds when the attacker gets a single try. Lower is better. |
| Gray Swan IPI (k = 10) | 17.4% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 10 tries and counts an attack as successful if any of them works. Lower is better. |
| Gray Swan IPI (k = 15) | 20.8% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 15 tries and counts an attack as successful if any of them works. Lower is better. |
| ProgramBench | 0.5% | [ProgramBench](https://programbench.com/), retrieved 2026-09-11 | The AI receives a working program and its documentation, then builds a replacement from scratch without the original source code, internet access or decompilation. The score is the percentage of 200 programs that pass every behavioral test. We record each model's best published mini-SWE-agent result, including higher reasoning efforts where available. Partial test-pass rates and almost-solved programs do not count toward this score. Equal scores share a rank here; the official board also uses partial progress to break ties. Higher is better. |
| SWE-Bench Pro | 58.6% | Lab | Can the AI fix real bugs in real software? It's handed actual problems from open-source projects and has to write code that genuinely solves them. Higher is better. |
| SWE-Bench Multilingual | 77.8% | Lab | Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. |
| DeepSWE 1.0 | 64.3% | Lab | Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. Higher is better. |
| Terminal-Bench 2.1 | 78.2% | Lab | Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better. |
| Terminal-Bench 2.0 | 82.7% | Lab | Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? (Version 2.0 of the test.) Higher is better. |
| Expert-SWE (Internal) | 73.1% | Lab | OpenAI's private set of expert-level software-engineering problems. Higher is better. |
| MCP Atlas | 75.3% | Lab | Can the AI chain together many tools and steps to complete one bigger task, rather than doing just a single thing? Higher is better. |
| Toolathlon | 55.6% | Lab | Tests how well the AI uses everyday real-world tools and apps to get things done. Higher is better. |
| BrowseComp | 84.4% | Lab | Can the AI browse the web and track down hard-to-find answers? Higher is better. |
| CyberGym | 81.8% | Lab | Tests the AI on cybersecurity challenges — finding and exploiting software weaknesses inside a safe sandbox. Higher is better. |
| Humanity's Last Exam (no tools) | 41.4% | Lab | Humanity's Last Exam — extremely hard expert questions across many subjects, written so you can't just look up the answer. “No tools” means the AI answers on its own. Higher is better. |
| Humanity's Last Exam (with tools) | 52.2% | Lab | Humanity's Last Exam — extremely hard expert questions across many subjects. “With tools” means the AI is allowed to search the web or run code while answering. Higher is better. |
| ARC-AGI-2 | 84.6% | Lab | Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better. |
| FrontierMath (Tier 1–3) | 51.7% | Lab | Very hard, research-level math problems. Tiers 1–3 are the (still extremely difficult) lower tiers. Higher is better. |
| FrontierMath (Tier 4) | 35.4% | Lab | Very hard, research-level math problems. Tier 4 is the hardest — close to what professional research mathematicians tackle. Higher is better. |
| GPQA Diamond | 93.6% | Lab | Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. |
| OSWorld-Verified | 78.7% | Lab | Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better. |
| Finance Agent v2 | 51.8% | Lab | Tests the AI on real financial-analysis work, like digging through reports and making sound decisions. Higher is better. |
| Harvey's Legal Agent Benchmark | 3.75% | Lab | Harvey's test of whether an AI agent can complete real legal work — drafting and reviewing documents, working with spreadsheets and presentations, and navigating files the way a lawyer's assistant would. Higher is better. |
| TaxEval v2 | 74.98% | Lab | A set of real tax questions created by Vals AI — can the AI give accurate answers about tax rules and filings? Higher is better. |
| MedScribe | 86.87% | Lab | Can the AI support doctors with their administrative work, like notes and paperwork? Created by Vals AI. Higher is better. |
| GDPval-AA | 1769 | Lab | Measures how well the AI does economically valuable knowledge work, judged against human experts. Shown as a rating (like a chess Elo) — higher is better. |
| GDPval-AA v2 | 1494 | Lab | economically valuable knowledge work (v2, re-based Elo) |
| GDPval (win/tie rate) | 84.9% | Lab | How often the AI's work matches or beats a human expert's on real knowledge-work tasks. Higher is better. |
| CharXiv Reasoning | 84.1% | Lab | Can the AI read and reason about complex charts and figures, not just text? Higher is better. |
| MMMU-Pro | 81.2% | Lab | A tougher version of MMMU — college-level questions that mix images, diagrams, and text together. Higher is better. |
| Blueprint-Bench 2 | 36.2% | Lab | Can the AI reason about space and layout — for example, understanding a floor plan or blueprint? Higher is better. |
| MRCR v2 (8-needle) (128k average) | 94.8% | Lab | Tests whether the AI can find specific details buried inside a very long document (around 128k tokens — roughly a long book). Higher is better. |

## About GPT-5.5

GPT-5.5, released April 23, 2026 alongside GPT-5.5-Pro, pushed OpenAI's context window past the million-token mark (1.05M) and posted the strongest long-context recall in its class — 94.8% on MRCR v2 at 128K. Its 84.6% on ARC-AGI-2 and 51.7% on FrontierMath Tiers 1–3 led all models at release on abstract reasoning and research mathematics.

On agentic work it scored 78.2% on Terminal-Bench 2.1, 78.7% on OSWorld-Verified, and 84.9% on GDPval win-rate — the highest real-world-task result of any model at the time. GPT-5.5 headlined OpenAI's lineup for two months until the three-model GPT-5.6 family (Sol, Terra, Luna) arrived on June 26, 2026.

## Questions and answers

### When was GPT-5.5 released?

GPT-5.5 was released by OpenAI on Thursday, Apr 23 2026.

### Who made GPT-5.5?

GPT-5.5 was built by OpenAI. Creators of ChatGPT and the GPT series of models. Pioneered large-scale language model research.

### How much does GPT-5.5 cost?

GPT-5.5 costs $5.00 per million input tokens and $30.00 per million output tokens through the OpenAI API. Cached input is $0.50 per million tokens. Rates are pay-as-you-go API prices verified against OpenAI's published pricing on August 18, 2026.

### What benchmark scores did GPT-5.5 get?

GPT-5.5 reports 33 tracked benchmark scores — BullshitBench v2: 47%; Gray Swan IPI (k = 1): 3%; Gray Swan IPI (k = 10): 17.4%; Gray Swan IPI (k = 15): 20.8%; ProgramBench: 0.5%; SWE-Bench Pro: 58.6%; SWE-Bench Multilingual: 77.8%; DeepSWE 1.0: 64.3%; Terminal-Bench 2.1: 78.2%; Terminal-Bench 2.0: 82.7%; Expert-SWE (Internal): 73.1%; MCP Atlas: 75.3%; Toolathlon: 55.6%; BrowseComp: 84.4%; CyberGym: 81.8%; Humanity's Last Exam (no tools): 41.4%; Humanity's Last Exam (with tools): 52.2%; ARC-AGI-2: 84.6%; FrontierMath (Tier 1–3): 51.7%; FrontierMath (Tier 4): 35.4%; GPQA Diamond: 93.6%; OSWorld-Verified: 78.7%; Finance Agent v2: 51.8%; Harvey's Legal Agent Benchmark: 3.75%; TaxEval v2: 74.98%; MedScribe: 86.87%; GDPval-AA: 1769; GDPval-AA v2: 1494; GDPval (win/tie rate): 84.9%; CharXiv Reasoning: 84.1%; MMMU-Pro: 81.2%; Blueprint-Bench 2: 36.2%; MRCR v2 (8-needle) (128k average): 94.8%. Tracked scores may come from lab reports or independent benchmarks; source details accompany the benchmark data. It holds the best score among all models tracked here on Expert-SWE (Internal), GDPval (win/tie rate) and Blueprint-Bench 2.

### What is the context window of GPT-5.5?

GPT-5.5 has a context window of 1.05M. That is the maximum amount of input plus output the model can hold in a single request.

### Is GPT-5.5 open source?

No. GPT-5.5 is a proprietary model. The weights are not published — it is available only through the provider's own API, apps, or partner platforms.

### What came before and after GPT-5.5?

OpenAI's previous tracked release was GPT-5.4 nano on Mar 17 2026, 37 days earlier. It was followed by GPT-5.5-Pro on Apr 23 2026.


---

Canonical page: https://aireleasetracker.com/model/openai/gpt-5.5
Full dataset: https://aireleasetracker.com/llms-full.txt · JSON: https://aireleasetracker.com/models.json
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.
