GPT-6.1 Sol hits Astra-level coding at a fifth of the API price
OpenAI's $2/$10 near-Astra model is live in Codex, Copilot and the API, and Alibaba's open-weight Qwen-Image-2.1 adds native RGBA editing.
Copy markdown
Astra coding, a fifth of the bill
GPT-6.1 Sol matches GPT-6 Astra on the DeepSWE 1.1 coding benchmark and lands within 2.1 points on OSWorld 2.0 computer-use — at $2/$10 per million tokens ($0.10 cached), a fifth of Astra's standard price. Call it as `gpt-6.1-sol` on the Responses API, with a ~1.05M-token context and 128K output.
Live in Codex, Copilot, and the API
Sol is already selectable for Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and in GitHub Copilot across VS Code, Visual Studio and JetBrains. The payoff is per-task cost: OpenAI clocks it at ~$5.47 vs ~$23.80 for Astra on Terminal-Bench Science — a 75%+ cut — with an Ultrafast tier up to 8x faster generation coming to Codex.
Qwen-Image-2.1 brings open-weight RGBA editing
Alibaba open-sourced a 7B unified image generation-and-editing model that outputs native transparency (RGBA), takes up to 10 reference images with mask-based local edits, and renders text at up to 2048×2048 — now the top trending model on Hugging Face. You can make transparent assets and do targeted edits locally, no cloud image API. Heads-up: it ships under the Qwen Research License, not a fully permissive one.