OpenAI shipped GPT-5.5 on April 23, 2026, just six weeks after GPT-5.4. Marketed as “a new class of intelligence for real work,” GPT-5.5 is a frontier model designed for multi-step coding, advanced computer use, and deep research. It’s available today in ChatGPT and Codex, with API access launching soon.
If you’re evaluating whether GPT-5.5 is worth adopting, this guide focuses on what’s new, how it differs from 5.4, key benchmarks, actionable usage steps, and practical caveats.
For hands-on guides, see:
To prepare for the API launch, get Apidog and pre-build your collection.
TL;DR
- GPT-5.5: OpenAI’s advanced coding/reasoning model, released April 23, 2026.
- Scores 88.7% on SWE-bench and 92.4% on MMLU; 60% fewer hallucinations vs. 5.4.
- Three variants: GPT-5.5 standard, GPT-5.5 Thinking (extended reasoning), GPT-5.5 Pro (highest accuracy).
- Available now in ChatGPT Plus, Pro, Business, Enterprise, Edu and Codex across all plans (including a temporary free window for Free and Go).
- API is staged—developers can use it via Codex sign-in path now; full API is rolling out soon.
- API pricing: $5/M input tokens, $30/M output tokens—double GPT-5.4, but more token-efficient.
What GPT-5.5 Actually Is
GPT-5.5 leads the GPT-5 family, above GPT-5.4, 5.4-mini, and 5.3. Codename “Spud,” official name GPT-5.5.
Variants:
- GPT-5.5 (default): Faster, sharper, and more token-efficient for most tasks.
- GPT-5.5 Thinking: Same model but with a larger reasoning budget—ideal for complex spreadsheets, dense research, and multi-file debugging. Capped at ~3,000 messages/week in ChatGPT.
- GPT-5.5 Pro: Highest-accuracy; for correctness-critical work. Only on Pro, Business, and Enterprise plans.
Key features:
Trained to plan, select tools, and self-check outputs. Expect fewer prompts, more accurate tables, and more clarifying questions instead of hallucinations.
What Changed from GPT-5.4
The six-week interval means targeted upgrades rather than a generational leap. Here’s the practical diff:
| Capability | GPT-5.4 | GPT-5.5 |
|---|---|---|
| SWE-bench | ~74 % | 88.7 % |
| MMLU | 91.1 % | 92.4 % |
| Hallucination rate | baseline | −60 % |
| Context window (API) | 1.05 M | 1 M (Codex: 400 K) |
| API input price | $2.50 / M | $5.00 / M |
| API output price | $15.00 / M | $30.00 / M |
| Computer use | Improving | Production-grade |
| Multi-step tool chains | Single-shot | Full autonomous loops |
SWE-bench is the headline metric. GPT-5.5’s 88.7% means it closes GitHub issues at a senior engineering level (per OpenAI). Test it on your codebase for real-world validation.
Pricing doubled, but token efficiency improves. Independent tests (The Decoder) show net cost rises ~20% overall, less on short-prompt workloads.
What It Is Good At
OpenAI targets four use cases:
- Agentic coding: Repo reading, file ops, running tests, iterative development. Powered by the SWE-bench metrics.
- Computer use: Drives browsers, shells, fills forms, scrapes data, recovers from intermediate errors.
- Deep research: Longer reasoning chains, better web search, and improved summarization. “Thinking” variant is optimized for this.
- Document/spreadsheet generation: Fewer layout errors, correct formulas, cleaner slides. Available in ChatGPT Plus/Business.
Not ideal for:
Cheap, high-volume tasks (classification, embeddings, bulk summarization)—use GPT-5.4-mini or 5.3 for better cost efficiency.
Availability Today
Here’s the access snapshot (as of April 23, 2026):
| Surface | Plan | Access |
|---|---|---|
| ChatGPT | Free / Go | GPT-5.3 default, no GPT-5.5 |
| ChatGPT | Plus | GPT-5.5 standard + Thinking (3,000/week) |
| ChatGPT | Pro / Business / Enterprise / Edu | Standard + Thinking + Pro |
| Codex | All plans (incl. Free/Go) | GPT-5.5 with 400K context; Free/Go on limited-time trial |
| API | Responses / Chat Completions | “Very soon”; not GA at launch |
Codex is the actionable path: GPT-5.5 is live in Codex, so you can use it from the CLI today—no API keys required.
Pricing in One Line
Budgeting? Here’s what to expect:
- GPT-5.5 API: $5/M input, $30/M output.
- GPT-5.5 Pro API: $30/M input, $180/M output (same as 5.4 Pro).
- Batch/Flex: Half standard rate.
- Priority: 2.5× standard rate.
- Codex CLI: Free on Plus, Pro, Business, Enterprise, Edu, Go, and temporarily Free (plan caps apply).
Full details and per-workload estimates in the GPT-5.5 pricing article.
How To Try It Today
Get started with minimal friction:
- Open ChatGPT (paid plan): Select GPT-5.5 from the model picker.
- Install Codex CLI:
npm install -g @openai/codex
# or
brew install codex
Then run codex and sign in with your ChatGPT account. Use /model gpt-5.5 to switch. Free/Go plans are included for a limited time.
Full walkthrough: How to use GPT-5.5 for free with Codex
- Pre-build for API rollout: The API is coming soon. Build your request collection now in Apidog so you’re ready when the model ID goes live. See the GPT-5.5 API guide for endpoint specs.
Safety and Red-Teaming
OpenAI has run GPT-5.5 through third-party cyber/bio risk tests ahead of launch. Safeguards, especially for offensive security tasks, are tighter with each model.
For developers:
- Expect stricter refusal behavior on dual-use code.
- The API rollout is staged while new safeguards are finalized.
- If you’re building consumer-facing agents, plan for more restrictive default policies than 5.4.
Should You Switch?
Decision matrix:
- Coding agents: Switch now. SWE-bench gains stack up, and Codex access is live.
- High-volume inference: Stick with GPT-5.4-mini for defaults; use GPT-5.5 only for hard cases.
- Consumer-facing products: Wait for API GA, then A/B test. The price jump is real; only switch if hallucination reduction matters for your users.
Full decision flow: pricing breakdown.
FAQ
Is GPT-5.5 available on the API?
Not yet for direct key-based calls (as of April 23, 2026). It’s live in Codex after ChatGPT sign-in, so early testers have access.
What is the context window?
1M tokens in ChatGPT and (soon) the API; 400K tokens in Codex CLI.
How do Thinking and Pro differ?
Thinking extends the reasoning budget on the standard model; Pro is a separate, higher-accuracy variant for critical tasks. Pro is available on Pro, Business, and Enterprise only.
Is GPT-5.5 free?
No, only paid ChatGPT plans. Codex offers temporary free access for Free and Go plans (rate limits apply). See the free guide for no-cost usage.
Can I still use GPT-5.4?
Yes. GPT-5.4 is still available and cheaper ($2.50/$15 per million tokens). For cost-sensitive pipelines, keep it as your default.
Top comments (0)