Artificial intelligence news.
Straight from the original announcements.

CodeRabbit puts GPT-6.2 Sol and Astra through coding review tests

November 19, 2026, 21:00By the AI News Bites Editorial TeamSource: CodeRabbit
CodeRabbit puts GPT-6.2 Sol and Astra through coding review tests
Image: CodeRabbit

CodeRabbit’s review of GPT-6.2 Sol and Astra found more known issues in its tests, but the models also produced substantially more comments and did not yet justify switching its coding defaults. The company said it will keep testing both models, with further work needed to separate model improvements from pipeline changes.

Key points
  • GPT-6.2 Astra caught 58 of 83 known issues, compared with 51 for GPT-6 Astra.
  • High-reasoning Sol 6.2 caught 60 of 81 known issues, compared with 58 for Sol 6.1.
  • Both GPT-6.2 configurations produced substantially more comments than their baselines.
  • Recorded Sol 6.2 high-reasoning cost per review rose to $1.146 from $0.772.
  • CodeRabbit will keep its current review defaults while testing both models further.

GPT-6.2 Astra caught more issues, with much more feedback

CodeRabbit tested GPT-6.2 Astra against GPT-6 Astra on a benchmark containing 83 known issues. The newer configuration caught 58 issues, or 69.9%, compared with 51, or 61.5%, for GPT-6 Astra.

That higher coverage came with a sharp increase in reported comments. GPT-6.2 Astra produced 847 comments, including 589 regular actionable comments and 258 findings in summary sections. The baseline produced 331 comments, although CodeRabbit said that total had not been reconciled to the same categories.

Actionable-comment benchmark precision fell from 21.74% for GPT-6 Astra to 12.22% for GPT-6.2 Astra. The newer configuration also produced 215 nitpick or assertive comments, compared with five for the baseline. Of the 589 regular actionable comments, 72 matched known issues. CodeRabbit said the other 517 comments could include useful findings outside the benchmark’s labels, but their value would need to be assessed individually.

Sol 6.2 delivered a small measured gain at higher cost

In a separate test, CodeRabbit evaluated Sol on 81 known issues across 42 pull requests. Sol 6.1 caught 58 issues, or 71.6%. Sol 6.2 with high reasoning caught 60, or 74.1%, while its low-reasoning configuration caught 59, or 72.8%.

The high-reasoning Sol 6.2 run produced 290 regular actionable comments, up from 187 for Sol 6.1. Benchmark precision declined from 35.3% to 25.9%. Recorded cost per review increased from $0.772 to $1.146, while median full-review time changed from 7m 49s to 7m 53s.

CodeRabbit cautioned that each configuration ran once and that different parts of the review pipeline used different model replacements. The results therefore do not establish a reliable model-only improvement. The reported figures are replay-run costs and full-pipeline timings, not production review costs or code-generation latency.

API prices fell, but completed review costs remain unclear

According to OpenAI’s pricing announcement, GPT-6.2 Astra’s standard API input price is $8 per million tokens, down from $10 for GPT-6 Astra. Cached input fell from $1 to $0.40 per million tokens, while standard output fell from $50 to $40 per million tokens.

CodeRabbit said those rates exclude long-context premiums and service-tier adjustments. Lower token prices also do not necessarily mean lower review costs, because completed reviews depend on delegated calls and retries. Only nine of 83 GPT-6.2 Astra cases had all costs recorded, compared with 58 cases for the baseline, so the test could not establish a reliable cost per completed review.

For timing, the GPT-6.2 Astra configuration’s median review time was 25 seconds longer than its baseline’s 5m 37s. CodeRabbit said that measures the full pipeline and cannot show that Astra itself is slower.

Hands-on coding results were mixed

CodeRabbit also used both models to build interfaces and work on projects, without running a matched code-generation benchmark. Astra often produced a useful first interface from a short prompt and selected layouts without waiting for clarification. On longer tasks, it returned to the original goal after detours and brought delegated investigations back into the work.

Sol 6.2 needed more direction in the company’s coding sessions. Its initial interface designs were not an improvement over Sol 6.1, and some implementations remained rough even with a reference. Small changes sometimes led to long pauses, while project setup repeatedly produced incorrect configurations that had to be checked.

CodeRabbit said it would keep its current review defaults while testing both models further. It plans to read the additional findings, retest issues each configuration missed and compare Astra using fixed routing before considering a switch.

Related news

GPT-6.2 Sol Matches Astra’s Accuracy at One-Quarter the Cost
ResearchNovember 19, 2026, 18:13

GPT-6.2 Sol Matches Astra’s Accuracy at One-Quarter the Cost

GPT-6.2 Sol delivered nearly the same auto-annotation accuracy as GPT-6.2 Astra in Ultralytics’ 36-dataset benchmark, while costing about one-quarter as much per 1,000…

GPT-6.2 Astra and Sol Put to the Test Building a 3D PyramidTop story
ModelsNovember 19, 2026, 16:07

GPT-6.2 Astra and Sol Put to the Test Building a 3D Pyramid

Kilo tested OpenAI’s GPT-6.2 Astra and GPT-6.2 Sol by asking both models to build an interactive 3D scene of workers constructing a pyramid. Both completed the project…

ModelsTop story
ModelsNovember 17, 2026, 18:45

OpenAI introduces GPT-6.2 Astra and Sol for work and science

OpenAI has introduced GPT-6.2 Astra for complex professional and scientific work, alongside GPT-6.2 Sol for everyday tasks. The company says both models deliver stronger…