Originally published at devtoolpicks.com

---

OpenAI shipped GPT-5.5 on April 23, 2026. The launch tweet hit 6.7 million views and 42,000 likes within hours. The internal codename before launch was "Spud," after the potato emoji OpenAI used to tease the release. The official name is GPT-5.5.

Here is what actually changed, what you get on which plan, and whether any of this shifts how you should be thinking about your AI coding setup right now.

What is GPT-5.5?

GPT-5.5 is OpenAI's new frontier model, released six weeks after GPT-5.4. OpenAI positions it as a model for agentic work: coding, computer use, web research, spreadsheets, multi-step tasks that require the model to keep going without a human stepping in every few minutes.

Three variants shipped at launch:

GPT-5.5 standard is the base model. Available in ChatGPT on Plus, Pro, Business, and Enterprise plans. Also in Codex on every plan including a temporary free window for Free and Go users.

GPT-5.5 Thinking extends the reasoning budget on the standard model. It works through harder problems before responding. Available on paid ChatGPT plans.

GPT-5.5 Pro is a separate higher-accuracy variant trained for correctness-critical tasks. Pro, Business, and Enterprise only.

What is not available yet: direct API key access. OpenAI says "coming very soon." For now, developers can reach GPT-5.5 through Codex after signing in with a ChatGPT account. The API pricing is confirmed at $5 per million input tokens and $30 per million output tokens.

What actually changed from GPT-5.4

Coding performance. This is the headline. GPT-5.5 scores 82.7% on Terminal-Bench 2.0, a benchmark testing complex command-line workflows that require planning and iterative tool use. Claude Opus 4.7 sits at 69.4% on the same benchmark. Gemini 3.1 Pro is at 68.5%. That is not a small gap.

On SWE-bench, which tests real software engineering tasks, GPT-5.5 scores 88.7%.

Hallucinations down 60%. OpenAI reports a 60% reduction in hallucinations compared to GPT-5.4. If you have built workflows around GPT-5.4 and found yourself double-checking outputs constantly, this is the number worth watching in practice.

Token efficiency. GPT-5.5 uses fewer tokens than GPT-5.4 to complete the same Codex tasks. That matters because the API price is double. The lower token count per task partially offsets the price increase, but how much depends on what you are doing.

Speed is unchanged. GPT-5.5 matches GPT-5.4's per-token latency in real-world serving. A smarter model at the same speed is unusual. Most capability jumps come with slower output.

Knowledge work benchmarks. On GDPval, which tests performance across 44 real professional jobs including finance, legal, and product management, GPT-5.5 matches or beats human professionals 84.9% of the time.

Context window. One million tokens in ChatGPT. 400,000 tokens in Codex CLI across every plan.

Access and pricing

In ChatGPT:

Codex:

GPT-5.5 is available in Codex on all plans, including a temporary free window for Free and Go users. Subject to per-plan weekly caps.

API pricing (when it arrives):

If you are already using the GPT-5.4 API at scale, the price jump is real and worth calculating against your current usage before switching.

The timing is interesting

GPT-5.5 did not launch into a quiet week for AI coding tools.

Anthropic published a confirmed performance drop in Claude Code on April 23, the same day as the GPT-5.5 launch. Developers reported slower, less reliable sessions on Claude Code compared to a few weeks earlier. Anthropic confirmed the drop was from engineering changes. No timeline given for when performance returns to the previous level.

This happened two days after Anthropic reversed a pricing page change that briefly removed Claude Code from the Pro plan entirely. That incident generated 32,500 posts on X in 24 hours and shook confidence in the platform for a lot of indie hackers who use Claude Code daily.

None of that makes GPT-5.5 automatically the better tool. But the context matters. If you have been less satisfied with Claude Code lately, the timing of this launch gives you a real alternative to evaluate rather than just waiting for Anthropic to fix things.

What this means for indie hackers

If you use ChatGPT Plus ($20/month) already: GPT-5.5 is available in your model picker right now. Switch to it and run your normal workflows. That is the quickest way to see whether the coding improvements are real for your specific use cases. No extra cost.

If you use Claude Code as your primary coding tool: The benchmark gap on Terminal-Bench 2.0 is significant enough to take seriously. GPT-5.5 through Codex CLI is now a direct competitor. Worth running both side by side on a real task for a day before deciding anything. The full Codex vs Claude Code comparison from last week covers the setup and workflow differences in detail.

If you rely on the API for production apps: The API is not available yet. Do not plan around it until OpenAI publishes a release date. When it does land, run the token math. Fewer tokens per task at double the price may work out cheaper in practice, or it may not, depending on your workload.

If you are on a tight monthly budget: GPT-5.5 on Plus at $20/month gives you the standard model. That is a reasonable way to test it. The meaningful capability jump is in the Pro variant, which requires a $200/month Pro plan. If you are not already on Pro, the standard model is the right starting point.

On the Claude Code performance drop: This is a short-term situation. Anthropic knows about it and will fix it. The solo developer Claude Code guide is still accurate for how the tool works at its best. If you have Claude Code in your workflow, the practical advice is to keep it running and add GPT-5.5 through Codex as a second option while Anthropic resolves the regression.

For the full context on the Pro plan situation and what it signals about Anthropic's pricing direction, the Claude Code Pro plan breakdown from yesterday covers it in detail.

FAQ

Is GPT-5.5 available for free?

On ChatGPT, no. Paid plans only (Plus, Pro, Business, Enterprise). On Codex CLI, there is a temporary free window for Free and Go users, subject to weekly rate limits. OpenAI has not said how long the free Codex access lasts.

How much will the GPT-5.5 API cost?

API pricing is confirmed at $5 per million input tokens and $30 per million output tokens. That is double the GPT-5.4 rate of $2.50 and $15. API access is not available yet as of April 24, 2026.

Is GPT-5.5 better than Claude Code for coding tasks?

On Terminal-Bench 2.0, GPT-5.5 scores 82.7% versus Claude Opus 4.7 at 69.4%. That is a meaningful difference on command-line and agentic coding tasks. Claude Code is also experiencing a confirmed performance drop right now that Anthropic has not yet resolved. Benchmarks and real-world performance are not always the same thing, but the gap is large enough to be worth testing directly.

What is GPT-5.5 Pro and do I need it?

GPT-5.5 Pro is a higher-accuracy variant trained for harder, correctness-critical tasks. It is available inside ChatGPT on Pro ($200/month), Business, and Enterprise plans. For most indie hackers and solo developers, the standard GPT-5.5 on Plus is the right starting point.

Does GPT-5.5 replace GPT-5.4?

Not yet. OpenAI has not deprecated GPT-5.4. It remains available and is cheaper at $2.50 per million input tokens versus $5 for GPT-5.5. If you are cost-sensitive and your current GPT-5.4 workflows are working well, there is no reason to switch immediately.

The bottom line

GPT-5.5 is a real upgrade, not a marketing bump. The coding benchmarks are the strongest evidence: an 82.7% score on Terminal-Bench 2.0 against Claude Opus 4.7 at 69.4% is a 13-point gap on a benchmark that directly tests the kind of work solo developers do.

The timing adds to the picture. Claude Code is in a rough week: a confirmed performance regression and a pricing scare that is still fresh. GPT-5.5 launching with strong coding numbers at the same moment is not coincidence, it is just how the AI market moves now.

If you are on ChatGPT Plus, switch the model picker to GPT-5.5 today and try it on a real task. That is the only honest way to know whether the improvement matters for your specific work. The benchmarks suggest it will.