OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracy
OpenAI introduced GPT-6.1 Sol on September 22, 2026 as the middle tier of its GPT‑6 family, available via the API and rolling out in ChatGPT Work and Codex for eligible Plus, Pro, Business, Enterprise and Edu plans. The company’s internal benchmark, reported by The New Stack on October 22, says the model achieved 86.6% accuracy on a low‑effort COBOL migration test and 97.9% when more reasoning…
Key points
- OpenAI’s internal test reports 86.6% low‑effort and 97.9% high‑effort COBOL migration accuracy for GPT‑6.1 Sol
- GPT‑6.1 Sol provides a 1,050,000‑token context window and five reasoning‑effort levels
- Pricing is $2 input/$10 output per million tokens, with cached input at $0.10 per million tokens
The model offers a 1,050,000‑token context window and five reasoning‑effort settings that trade speed for accuracy. Pricing remains $2.00 per million input tokens, $10.00 per million output tokens, with cached input at $0.10 per million tokens. Compared with GPT‑6 Astra, OpenAI claims GPT‑6.1 Sol delivers near‑flagship performance at lower cost, while GPT‑6 Luna stays the low‑cost, high‑volume option.
For banks, insurers and government IT shops that still run legacy COBOL systems, the reported accuracy could make modernization projects more affordable, but the lack of external verification means pilots are advisable before large‑scale adoption.
Model pages: GPT-6 Sol → · GPT-5.6 Sol → · GPT-6 Astra →
The story so far
6 episodes →- OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracythis story
OpenAI’s GPT-6.1 Sol shows dramatic improvements in legacy code migration including COBOL
online-tech-tips.com · 24 October 2026
Loading the full article…
This text was published by online-tech-tips.com and written by Aseem Kishore. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Nanbeige releases Nanbeige4.3-33B teacher model for lightweight agents · 1 src
- Moonshot AI releases Kimi K3.1 Pro, a 2.8T open-weight model closing gap with proprietary leaders · 8 src
- Google Cloud switches AlphaEvolve default to Gemini 4 Argon and adds OpenAI GPT-6 models · 1 src
- Anthropic investigates elevated errors across five Claude models · 1 src
- Llama.cpp PR adds DeepSeek-V4.1-Flash support with review fixes · 1 src
Comments
via GitHub Discussions