GPT-6.1 Sol
| Maker | OpenAI |
|---|---|
| Family | GPT Sol |
| Released | 2026-09-29 |
| Context | 1.1M |
| Max output | 128K |
| Open weights | Closed |
|---|---|
| Modalities | text, image, pdf → text |
| Input / 1M | $2.00 |
| Output / 1M | $10.00 |
| Cached input / 1M | $0.10 |
| Above 272K in | $4.00 / $15.00 |
|---|---|
| Review call | $0.0766 |
| Reasoning efforts | low, medium, high, xhigh |
| Hosts | OpenAI, OpenRouter |
| Model id | gpt-6.1-sol |
- Rank
- #1=
- of 17
- Bugs caught
- 89%
- 80%–97%
- False alarms
- 0.9
- per clean PR
- Cost / review
- $0.0401
- Cost / bug
- $0.0514
- Time
- 28s
- per review
If you want one default seat, GPT-6.1 Sol makes a strong case: 89% [80%–97%] recall for $0.0401 a review. It is measurably above 7 of 17 ranked models on recall, and 0 are measurably above it.
The trade that matters is against Claude Opus 5.5. Their recall can't be separated here, yet an Opus review cost 3.8× as much, and Sol wrote shorter reviews. Opus measurably beats 9 ranked models and Sol beats 7, but neither is measurably above the other.
Sol averaged 0.9 [0.4–1.5] false alarms per clean PR (Opus: 1.7 [1.1–2.3]), and 95% of its catches point at the exact lines of the defect. A reviewer that cites the right line saves reading time even when the comment is short.
Costs to know about:
- At medium effort a review took 28s. Claude Sonnet 5.5 took 11s and lists at $2.00 / $10.00 per 1M tokens (Sol: $2.00 / $10.00), but its measured review cost 1.4× Sol's, and the run can't separate their recall.
- OpenAI makes one of the three judges, and its vote is dropped for Sol.
- "Refactor: MudDebouncedInput, ThrottleDispatcher and DebounceDispatcher" was caught in some runs and missed in others, so a single pass can still slip.
Read the results post for how it was first measured.
Figures follow the latest benchmark run and price check. Text checked Nov 5, 2026.
| Host | Model id | Input / 1M | Output / 1M |
|---|---|---|---|
| OpenAI | gpt-6.1-sol | $2.00 | $10.00 |
Source: OpenAI pricing, Oct 16, 2026
By kind of bug
- Logic 14
- 100%
- Security 9
- 89%
- Concurrency 8
- 71%
- Error handling 5
- 80%
- API misuse 3
- 93%
- Resource handling 3
- 100%
Bug by bug
- Known bugs
- 42
- Caught every run
- 36
- Caught sometimes
- 2
- Missed
- 4
- Clean PRs
- 15
- False alarms on them
- 16
| Result | Pull request | Language | Kind |
|---|---|---|---|
| Caught sometimes4/5 | 4.x: Enhance Streamable/Streamer and DisposableContainer APIsReactiveX/RxJava | Java | API misuse |
| Caught sometimes2/3 | Refactor: MudDebouncedInput, ThrottleDispatcher and DebounceDispatcherMudBlazor/MudBlazor | C# | Concurrency |
| Missed | rpc : cache and reuse compute graphsggml-org/llama.cpp | C++ | Security |
| Missed | perf(gitdiff): optimize inline diff highlighting using cachego-gitea/gitea | Go | Error handling |
| Missed | Store cached child els in Attributes, not fieldjhy/jsoup | Java | Concurrency |
| Missed | fix(server): pass maxDurationMs timeout signal to subscription handlerstrpc/trpc | TypeScript | Concurrency |
| Version | Released | Input / 1M | Output / 1M | Context | Rank | Bugs caught |
|---|---|---|---|---|---|---|
| GPT-6 Sol | 2026-09-22 | $2.00 | $10.00 | 1.1M | – | – |
| GPT-5.6 Sol | 2026-07-09 | $4.00 | $20.00 | 1.1M | #1 | 72% |
| Model | Maker | Input / 1M | Output / 1M | Review call | Context | Bugs caught |
|---|---|---|---|---|---|---|
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $0.0766 | 1M | 77% |
| Gemini 4 Argon | $2.00 | $10.00 | $0.0766 | 1M | 82% | |
| Kimi K2.7 Code Highspeed | Moonshot | $1.90 | $8.00 | $0.0691 | 262.1K | – |
| Grok 4.7 | xAI | $2.00 | $6.00 | $0.0669 | 500K | – |
Updated Nov 5, 2026