OpenAI released GPT-5.5 on April 23, 2026, and it is not a minor bump. Internally codenamed "Spud," the model retakes the frontier lead in publicly available LLMs, beating Claude Opus 4.7, Gemini 3.1 Pro, and even narrowing the gap with Anthropic's private Claude Mythos Preview on key benchmarks. Here is what you need to know.
What Changed from GPT-5.4 to GPT-5.5
GPT-5.5 is not just bigger. It is more efficient. According to OpenAI, it matches GPT-5.4's per-token latency in real-world serving while delivering higher intelligence across the board. It also uses fewer tokens to complete the same tasks — particularly in Codex, where it solves more issues in a single pass.
The headline improvements land in three areas:
- Agentic coding: Terminal-Bench 2.0 jumps from 75.1% to 82.7%, and Expert-SWE (OpenAI's internal eval for tasks averaging 20 human hours) goes from 68.5% to 73.1%.
- Computer use: OSWorld-Verified climbs to 78.7%, and the model can operate software, move between tools, and handle messy multi-step workflows with less guidance.
- Research and reasoning: FrontierMath Tier 1–3 reaches 51.7% (up from 47.6%), and BrowseComp hits 84.4%.
Benchmark Comparison
| Benchmark | GPT-5.5 | GPT-5.4 | GPT-5.5 Pro | GPT-5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
|---|---|---|---|---|---|---|
| Terminal-Bench 2.0 | 82.7% | 75.1% | — | — | 69.4% | 68.5% |
| Expert-SWE (Internal) | 73.1% | 68.5% | — | — | — | — |
| GDPval (wins/ties) | 84.9% | 83.0% | 82.3% | 82.0% | 80.3% | 67.3% |
| OSWorld-Verified | 78.7% | 75.0% | — | — | 78.0% | — |
| BrowseComp | 84.4% | 82.7% | 90.1% | 89.3% | 79.3% | 85.9% |
| FrontierMath T1–3 | 51.7% | 47.6% | 52.4% | 50.0% | 43.8% | 36.9% |
| FrontierMath T4 | 35.4% | 27.1% | 39.6% | 38.0% | 22.9% | 16.7% |
| CyberGym | 81.8% | 79.0% | — | — | 73.1% | — |
GPT-5.5 Pro leads on BrowseComp (90.1%) and FrontierMath Tier 4 (39.6%), making it the pick for tasks that demand deep precision.
What Developers Are Saying
Early testers report a noticeable step change. Dan Shipper (CEO of Every) tested GPT-5.5 on a production bug that had stumped GPT-5.4 and his best engineer eventually had to rewrite. GPT-5.5 produced the same rewrite on its first attempt. He called it "the first coding model I've used that has serious conceptual clarity."
Pietro Schirano (CEO of MagicPath) saw GPT-5.5 merge a branch with hundreds of frontend changes into a substantially changed main branch — resolving everything in a single shot in about 20 minutes.
Senior engineers at NVIDIA and other companies noted that GPT-5.5 catches issues proactively, predicts testing and review needs without prompting, and produces plans that need less implementation correction than GPT-5.4.
Pricing and Availability
GPT-5.5 is available now for ChatGPT Plus ($20/month), Pro ($100–200/month), Business, and Enterprise users. GPT-5.5 Pro requires a Pro or higher subscription.
API access launched April 24 for both GPT-5.5 and GPT-5.5 Pro. According to OpenAI, GPT-5.5 delivers state-of-the-art coding performance at roughly half the cost of competitive frontier coding models on the Artificial Analysis Intelligence Index.
GPT-5.4 remains available in the API at a lower price point if you do not need the latest capabilities.
Should You Upgrade?
If you are building agentic tools, doing complex code generation, or working with multi-step workflows, GPT-5.5 is a clear upgrade. The combination of higher benchmark scores with lower token usage and matched latency means better results for less money.
If your use case is straightforward text generation or simple completions, GPT-5.4 at its reduced price is still a solid choice.
What This Means for the AI Landscape
GPT-5.5 pushes the frontier forward in agentic coding and computer use, areas where Anthropic's Claude Opus 4.7 and Google's Gemini 3.1 Pro had been competitive. The gap on Terminal-Bench 2.0 is significant: 82.7% vs 69.4% vs 68.5%. The model's ability to handle messy, ambiguous tasks with minimal guidance is the key differentiator.
The AI model space in April 2026 is moving fast. OpenAI's incremental versioning (5.4 → 5.5) belies what is a substantial capability jump. As Fortune noted, "AI model launches are starting to look like software updates" — but these updates carry real performance improvements.
We are tracking GPT-5.5 and all major model releases in our model directory. Check the full comparison at q4km.ai.