GPT-5.5
Released
GPT-5.5 is an AI model released by OpenAI on Thursday, Apr 23 2026, 37 days after GPT-5.4 nano. It has a 1.05M token context window. Benchmark results (shown below) cover Expert-SWE (Internal), GDPval (win/tie rate), Blueprint-Bench 2, BullshitBench v2, Gray Swan IPI, ProgramBench, and 23 more.
Get OpenAI releases and AI news.
One email, every other Monday.
API pricing
- InputWhat you pay for everything you send the model — your question, plus any documents or earlier conversation you include with it.
- $5.00
- Cached inputA reduced rate for text you send over and over. If every request starts with the same instructions or the same document, the provider keeps a copy ready and charges less to read it again.
- $0.50
- OutputWhat you pay for the text the model writes back. It is normally the dearer half: producing an answer costs more than reading one.
- $30.00
Available from
| Azure | — | $5.00 | $30.00 | 1M |
|---|---|---|---|---|
| Azure | eu | $5.50 | $33.00 | 1M |
| Azure | us | $5.50 | $33.00 | 1M |
| Amazon Bedrock | us-east-1 | $5.50 | $33.00 | 1M |
Benchmarks
Coding
Expert-SWE (Internal)Software engineering — OpenAI's private set of expert-level software-engineering problems. Higher is better.
73.1%
#1Best published Expert-SWE score of all tracked models
ProgramBenchProgram reconstruction — 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.
0.5%
via ProgramBench
#3 of 17Best: Claude Opus 5 · 4.5%
SWE-Bench ProAgentic coding — 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.
58.6%
#14 of 23Best: Claude Fable 5.1 · 81.2%
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better.
77.8%
#9 of 14Best: Claude Opus 5 · 89.5%
DeepSWE 1.0Agentic coding — 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.
64.3%
#3 of 6Best: Kimi K3 · 67.5%
Terminal & CLI
Terminal-Bench 2.1Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better.
78.2%
#20 of 30Best: DeepSeek-V4.1-Flash · 90.6%
Terminal-Bench 2.0Agentic terminal coding — 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.
82.7%
#2 of 14Best: Claude Mythos 5 · 88%
Agentic & tool use
MCP AtlasMulti-step tool use — Can the AI chain together many tools and steps to complete one bigger task, rather than doing just a single thing? Higher is better.
75.3%
#9 of 11Best: Muse Spark 1.1 · 88.1%
ToolathlonGeneral tool use — Tests how well the AI uses everyday real-world tools and apps to get things done. Higher is better.
55.6%
#2 of 5Best: Gemini 3.5 Flash · 56.5%
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better.
84.4%
#12 of 28Best: GPT-5.6 Sol · 92.2%
CyberGymCybersecurity — Tests the AI on cybersecurity challenges — finding and exploiting software weaknesses inside a safe sandbox. Higher is better.
81.8%
#6 of 10Best: DeepSeek-V4.1-Flash · 88.1%
OSWorld-VerifiedAgentic computer use — Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better.
78.7%
#9 of 28Best: Qwen3.8-Max · 86.1%
Reasoning & science
Humanity's Last ExamMultidisciplinary reasoning — 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.
41.4%
no tools
52.2%
with tools
no tools
#9 of 22Best: Claude Fable 5.1 · 60.9%
with tools
#19 of 42Best: Claude Fable 5.1 · 65%
ARC-AGI-2Abstract reasoning — Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better.
84.6%
#5 of 21Best: GPT-6 Astra · 95%
FrontierMathAdvanced math — Very hard, research-level math problems. Tiers 1–3 are the (still extremely difficult) lower tiers. Higher is better.
51.7%
Tier 1–3
35.4%
Tier 4
Tier 1–3
#2 of 6Best: GPT-5.5-Pro · 52.4%
Tier 4
#3 of 6Best: GPT-5.5-Pro · 39.6%
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
93.6%
#5 of 59Best: GPT-6 Astra · 96%
Knowledge work
GDPval (win/tie rate)Knowledge work — How often the AI's work matches or beats a human expert's on real knowledge-work tasks. Higher is better.
84.9%
#1Best published GDPval WR score of all tracked models
GDPval-AAKnowledge work — 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.
1769
#3 of 9Best: Claude Fable 5 · 1932
GDPval-AA v2Knowledge work — economically valuable knowledge work (v2, re-based Elo)
1494
#15 of 20Best: Claude Opus 5 · 1861
Finance
Finance Agent v2Agentic financial analysis — Tests the AI on real financial-analysis work, like digging through reports and making sound decisions. Higher is better.
51.8%
#5 of 9Best: Gemini 3.8 Flash · 61.4%
Legal & tax
Harvey's Legal Agent BenchmarkAgentic legal work — 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.
3.75%
#8 of 8Best: Muse Spark 1.1 · 20%
TaxEval v2Tax questions — A set of real tax questions created by Vals AI — can the AI give accurate answers about tax rules and filings? Higher is better.
74.98%
#4 of 4Best: Muse Spark 1.1 · 79.72%
Healthcare
MedScribeMedical admin work — Can the AI support doctors with their administrative work, like notes and paperwork? Created by Vals AI. Higher is better.
86.87%
#4 of 5Best: Muse Spark 1.1 · 88.89%
Multimodal
Blueprint-Bench 2Spatial reasoning — Can the AI reason about space and layout — for example, understanding a floor plan or blueprint? Higher is better.
36.2%
#1Best published Blueprint score of all tracked models
CharXiv ReasoningChart reasoning — Can the AI read and reason about complex charts and figures, not just text? Higher is better.
84.1%
#9 of 16Best: Muse Spark · 88.9%
MMMU-ProMultimodal reasoning — A tougher version of MMMU — college-level questions that mix images, diagrams, and text together. Higher is better.
81.2%
#5 of 12Best: Gemini 3.5 Flash · 83.6%
Long context
MRCR v2 (8-needle)Long context — Tests whether the AI can find specific details buried inside a very long document (around 128k tokens — roughly a long book). Higher is better.
94.8%
128k average
#2 of 7Best: Gemini 3.7 Flash · 97%
Robustness
BullshitBench v2Nonsense detection — 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.
47%
#33 of 77Best: Claude Opus 4.8 · 95%
Gray Swan IPIPrompt injection robustness — 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.
3%
k = 1
17.4%
k = 10
20.8%
k = 15
k = 1
#6 of 13Best: Claude Opus 5 · 0.2%
k = 10
#7 of 13Best: Claude Opus 5 · 1.6%
k = 15
#7 of 13Best: Claude Opus 5 · 2%
Source: ProgramBench, retrieved 11 September 2026. Every other score here is the figure the lab published at launch.
About
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.
Compare GPT-5.5 with
Suggested comparisons
Frequently asked questions
GPT-5.5 was released by OpenAI on Thursday, Apr 23 2026.
All OpenAI releases
44 tracked2026
14 releasesGPT-6 Astra
Sep 3 2026
GPT-5.6-Cyber
Aug 10 2026
GPT-5.6 Sol
Jun 26 2026
GPT-5.6 Terra
Jun 26 2026
GPT-5.6 Luna
Jun 26 2026
GPT-5.5-Cyber
Jun 22 2026
GPT-5.5
Apr 23 2026
GPT-5.5-Pro
Apr 23 2026
GPT-5.4 mini
Mar 17 2026
GPT-5.4 nano
Mar 17 2026
GPT-5.4
Mar 5 2026
GPT-5.4-Pro
Mar 5 2026
GPT-5.3-Codex-Spark
Feb 12 2026
GPT-5.3-Codex
Feb 5 2026
2025
22 releasesGPT-5.2
Dec 11 2025
GPT-5.1-Codex-Max
Nov 19 2025
GPT-5.1
Nov 12 2025
GPT-5-Codex-Mini
Nov 7 2025
GPT-5-Codex
Sep 15 2025
GPT-5
Aug 7 2025
GPT-5 mini
Aug 7 2025
GPT-5 nano
Aug 7 2025
GPT-5 Pro
Aug 7 2025
gpt-oss-120b
Aug 5 2025
gpt-oss-20b
Aug 5 2025
o3-pro
Jun 10 2025
o3
Apr 16 2025
o4-mini
Apr 16 2025
o4-mini-high
Apr 16 2025
GPT-4.1
Apr 14 2025
GPT-4.1 mini
Apr 14 2025
GPT-4.1 nano
Apr 14 2025
o1-pro
Mar 19 2025
GPT-4.5
Feb 27 2025
o3-mini
Jan 31 2025
o3-mini-high
Jan 31 2025