# GPT-5.5 reasoning curve evidence

Stet publishes a public search API over its indexed pages; the guide is at https://www.stet.sh/agents/agent-access.

Source: https://www.stet.sh/benchmarks/gpt-55-codex-graphql-reasoning-curve
Source observed: 2026-11-08T19:52:04.660Z
Last checked: 2026-11-08T19:52:04.66+00:00
Indexed: 2026-11-08T20:48:05.913+00:00

---

# GPT-5.5 reasoning curve evidence

We ran GPT-5.5 Codex at low, medium, high, and xhigh reasoning effort on May 7, 2026, across 26 matched real coding tasks in `graphql-go-tools`. This evidence page compares test results, equivalence, code-review quality, footprint risk, craft and discipline, cost, and duration across the four effort levels.

> On May 7, 2026, equivalence rose from 4/26 at low effort to 23/26 at xhigh effort, and code-review pass rose from 3/26 to 18/26. Test pass was not monotonic: 21 at low, 21 at medium, 25 at high, and 24 at xhigh.

## Benchmark details

| Label | Value |
| --- | --- |
| Tasks | 26 real coding tasks |
| Repository | `graphql-go-tools` |
| Judge | GPT-5.4 |
| Date | 2026-05-07 |
| Harness | OpenAI Codex CLI 0.128.0 |
| Caveat | Single seed per task; 26-task slice; no human grader calibration on this set. |

## Results table

| Metric | Low | Medium | High | Xhigh |
| --- | ---: | ---: | ---: | ---: |
| Tests pass | 21 / 26 | 21 / 26 | 25 / 26 | 24 / 26 |
| Equivalence | 4 / 26 | 11 / 26 | 18 / 26 | 23 / 26 |
| Code-review pass | 3 / 26 | 5 / 26 | 10 / 26 | 18 / 26 |
| Footprint risk | 0.200 | 0.268 | 0.314 | 0.365 |
| Craft / discipline avg | 2.311 | 2.604 | 2.736 | 3.071 |
| Cost per task (mean) | $2.65 | $3.13 | $4.49 | $9.77 |
| Mean agent duration | 286.9s | 411.0s | 579.0s | 753.3s |

## Low vs Medium

Tests tied at 21/26, but medium effort increased equivalence from 4/26 to 11/26 and craft/discipline from 2.311 to 2.604. Tests alone would not show this reasoning-effort difference.

## Medium vs High

High was the cleanest practical upgrade: tests increased from 21 to 25, equivalence from 11 to 18, and review pass from 5 to 10. Cost increased 1.43x, while integration details were correct more often.

## High vs Xhigh

Xhigh increased equivalence from 18 to 23 and review pass from 10 to 18 at 2.18x the cost. Tests fell from 25 to 24, and footprint risk widened. We characterize xhigh as a quality mode rather than a default.

## Related resources

- [Source writeup](https://www.stet.sh/blog/gpt-55-codex-graphql-reasoning-curve)
- [Scoring and validation details](https://www.stet.sh/methodology)
- [Run this coding-agent benchmark on your codebase](https://www.stet.sh/private)

---

[Read this page’s summary](https://www.stet.sh/agents/api/public/v1/page-tldr?url=https%3A%2F%2Fwww.stet.sh%2Fbenchmarks%2Fgpt-55-codex-graphql-reasoning-curve&trace=t_5da45423e484)

## Ask a follow-up about this page

Need a more specific answer? Ask the exact question you have. It returns a fresh, cited answer from Stet’s knowledge base. No account or API key required.

Open this URL with your own question, URL-encoded. Include your question in q. All research inputs are optional; no account or setup.

`GET https://www.stet.sh/agents/agent/faq?q={question}&trace=t_5da45423e484`

[Tool guide and examples](https://www.stet.sh/agents/agent-access) · [OpenAPI](https://www.stet.sh/agents/api/public/v1/openapi.json) · [Page TLDR form](https://www.stet.sh/agents/agent-tools/page-tldr?url=https%3A%2F%2Fwww.stet.sh%2Fbenchmarks%2Fgpt-55-codex-graphql-reasoning-curve). Opening a form does not execute a tool.