GPT-5.5
OpenAI's flagship for coding and professional work — "a new class of intelligence." Intelligence Index 60 (#3 overall), GDPval 84.9%, Terminal-Bench 2.0 82.7%, and a 1M-token context window. Released April 23, 2026.
Model Specs
- Released
- Apr 2026
- Context window
- 1.1M tokens
- Max output
- 128K tokens
- Capabilities
- reasoningmultimodalfunction-callinglong-context
- Modalities
- textvision
About this model
GPT-5.5 is OpenAI's flagship model, released April 23, 2026 and described by the company as "a new class of intelligence for coding and professional work." Greg Brockman summarized the generational change against GPT-5.4: a faster, sharper thinker that reaches better answers in fewer tokens. On the Artificial Analysis Intelligence Index it scores 60 at the xhigh effort tier — the #3 model overall — with a measured context window of roughly 920K usable tokens out of an official 1,050,000.
The headline results target real work rather than puzzle benchmarks. On **GDPval** — OpenAI's evaluation built from tasks drawn from real occupations — GPT-5.5 scores 84.9%, a major jump over GPT-5.2's 70.9%. On Terminal-Bench 2.0 (agentic terminal workflows) it posts 82.7%, and on SWE-Bench Pro 58.6%. The model ships in multiple forms: the API reasoning model with xhigh/high effort tiers, GPT-5.5 Instant (the fast ChatGPT default), and GPT-5.5 Pro for the highest-stakes work. It retains the full agentic stack — tool use, computer use, structured outputs, and deep Codex integration.
Raw API pricing is $5/M input and $30/M output — half the price of Claude Fable 5, which is its main rival at the frontier. Fable 5 leads on raw intelligence (64.9 vs 60) and hard coding (SWE-Bench Pro 80.3% vs 58.6%); GPT-5.5 counters with lower cost, a mature agentic toolchain, and OpenAI ecosystem depth. Reach for GPT-5.5 when you want frontier-class reasoning for professional and agentic work at a sane price point — and note that reasoning depth comes with latency: ~103 seconds to first token at xhigh effort, streaming ~51 tokens/sec after.
Key Strengths
Top-3 frontier intelligence
Artificial Analysis Intelligence Index of 60 at xhigh effort — #3 among all models, behind only Claude Fable 5. The high-effort tier scores 59 for faster turnaround at near-identical quality.
Best-in-class real-work performance
84.9% on GDPval — OpenAI's benchmark of tasks from real occupations — up from 70.9% for GPT-5.2. The model is explicitly tuned for professional deliverables, not just academic evals.
Agentic terminal mastery
82.7% on Terminal-Bench 2.0 — among the strongest published scores for autonomous terminal workflows: running builds, debugging, managing environments end-to-end.
1M-token context window
Official 1,050,000-token input window (AA measures ~920K usable) with 128K output tokens — full codebases, document sets, and long agent histories in one request.
Half the price of its frontier rival
$5/M input and $30/M output raw — exactly half of Claude Fable 5's $10/$50. For frontier-adjacent workloads at volume, the cost difference compounds fast.
Effort-tiered reasoning
Choose xhigh for maximum quality or high for faster, cheaper runs at one Index point lower. The same model spans deep research and production API duty.
How it compares
GPT-5.5 sits in the frontier tier alongside Claude Fable 5, with GPT-5.2 and fast models below it. Compare on intelligence, price, and ecosystem.
| vs. Model | Verdict | Outcome |
|---|---|---|
| GPT-5.2 | GPT-5.5 is the clear generational upgrade: Intelligence Index 60 vs 51, GDPval 84.9% vs 70.9%, and ~1M context vs 400K. GPT-5.2 remains cheaper raw ($1.75/$14 vs $5/$30) and is still excellent for everyday flagship duty — but for hard professional and agentic work, GPT-5.5 wins outright. | Wins most cases |
| Claude Fable 5 | Fable 5 leads on intelligence (64.9 vs 60), hard coding (SWE-Bench Pro 80.3% vs 58.6%), and GDPval-AA Elo (1932 vs 1769). GPT-5.5 is half the price ($5/$30 vs $10/$50) with a mature agentic/computer-use stack and Codex integration. Budget-sensitive frontier work favors GPT-5.5; the absolute hardest problems favor Fable 5. | Other wins |
| Gemini 3.5 Flash | Different tiers: Gemini 3.5 Flash (Index 55) is a speed-and-cost play — 152 t/s, 18.7s first token, $1.50/$9 — while GPT-5.5 (Index 60) is deeper but slower and ~3x the price. Flash for interactive and high-volume agents; GPT-5.5 when reasoning depth is worth the wait. | Depends |
Pros
- #3 on the Artificial Analysis Intelligence Index (60 at xhigh effort)
- GDPval 84.9% — best published score on real-occupation tasks
- Terminal-Bench 2.0 82.7% for agentic terminal workflows
- ~1M-token context (1,050,000 official) with 128K output
- Half the raw price of Claude Fable 5 ($5/$30 vs $10/$50)
- Effort tiers (xhigh/high) trade speed against depth on the same model
- Recent knowledge cutoff (December 1, 2025)
Things to consider
- Slow time-to-first-token at xhigh effort (~103s) — not for latency-sensitive UX
- Trails Claude Fable 5 on raw intelligence and hard coding benchmarks
- Verbose reasoning consumes output tokens at the $30/M rate
- GPQA Diamond / AIME scores not published in accessible sources at launch
- Premium over GPT-5.2 only pays off on genuinely hard tasks
Best use cases
Professional deliverables
Reports, analyses, plans, and client-grade documents — the GDPval 84.9% result reflects exactly this category of real-occupation tasks.
Agentic coding & Codex workflows
Terminal-Bench 82.7% and deep Codex integration make GPT-5.5 a natural engine for autonomous coding agents that build, test, and iterate.
Long-context analysis
Load entire codebases or document collections into the ~1M window for architecture reviews, contract analysis, or cross-document synthesis.
Computer-use automation
GPT-5.5 retains OpenAI's computer-use capability for agents that operate real interfaces — form filling, multi-app workflows, browser tasks.
High-volume reasoning at controlled cost
The high-effort tier delivers near-flagship quality faster and cheaper — a practical default for production pipelines that GPT-5.2 used to serve.
Research synthesis
Frontier reasoning plus a December 2025 knowledge cutoff make it a strong base for literature reviews and technical research grounded in recent knowledge.
How to use it on Renas AI
- 1
Step 1
Open AI Chat on Renas
Navigate to AI Chat in the Renas dashboard and pick the most capable OpenAI model available to your plan from the model selector.
- 2
Step 2
Frame the task like a brief
GPT-5.5 excels at professional work — give it the deliverable, audience, and constraints the way you'd brief a colleague, and include all relevant source material.
- 3
Step 3
Use long context deliberately
Paste whole files and documents rather than summaries. The ~1M-token window keeps everything in scope, and prompt-level completeness is what unlocks the GDPval-class output quality.
- 4
Step 4
Expect deliberation on hard tasks
At high reasoning effort the model thinks before it streams (~103s to first token on demanding requests). For quick interactions, use a lighter model and reserve GPT-5.5 for work that matters.
Pricing
Pricing on Renas AI
Pay-as-you-go credits, no API keys, no rate limits.
~66,667 words on a 10,000-credit Spark plan
Frequently asked questions
Other OpenAI models
Other text models on Renas AI
Frontier OpenAI intelligence on Renas
Access leading OpenAI models with your Renas AI subscription credits — no API key, no setup, no per-seat fees.
Try OpenAI models on Renas