OpenAI Launches GPT-6.1 Sol: Near-Astra Intelligence for Coding and Professional Work at One-Fifth the Price

OpenAI launched GPT-6.1 Sol, an upgraded version of GPT-6 Sol built to deliver near-Astra-level performance on coding, computer use, and professional work at a fraction of Astra’s price.

Details

• Vs. GPT-6 Astra: matches Astra on coding and professional tasks at roughly one-fifth the cost, approaches Astra on scientific research at over 75% lower expense, and stays within 1.9 percentage points of Astra’s factuality error rate
• Vs. GPT-6 Sol: gains 6.4 points on complex software engineering, 4.8 points on multi-step business automation, and 7 points on computer-use tasks at half the cost, while more than doubling Sol’s scientific research performance
• Benchmarks: matches Astra on DeepSWE v1.1 (coding); beats Opus 5.5 at lower cost on GDPval.pdf (document analysis); leads on AutomationBench (workflows) and OSWorld 2.0 (computer interaction); doubles prior Sol’s score on Terminal-Bench Science 0.1; cuts factuality errors by about 32% at low reasoning effort
• Pricing (API): $2 per million input tokens, $0.10 per million cached input tokens (a 95% discount), $10 per million output tokens
• Availability: rolling out now in ChatGPT Work, Codex, and the API for all usage tiers; an “Ultrafast” variant is coming soon

What happened next

GPT-6.1 Sol launched alongside GitHub Copilot’s own rollout of the model and OpenAI’s “dots” proactive-assistant product at DevDay 2026, extending the same efficiency-over-raw-power pattern seen in Claude Sonnet 5.5 and Grok 4.7 — each chasing Astra/Opus-tier quality at a steep cost discount rather than a new capability ceiling.