GPT-6.1 Sol: Near-Flagship Intelligence for 80% Less

OpenAI's new GPT-6.1 Sol model offers intelligence comparable to the flagship GPT-6 Astra at 20% of the price. It features significant performance improvements in coding, scientific research, and business automation while drastically lowering the cost of cached inputs. The model is now available for API developers and professional ChatGPT users, with an 'Ultrafast' version coming soon.

Key Points

• GPT-6.1 Sol delivers near-Astra intelligence at one-fifth the price of the flagship model.
• Cached input costs have been reduced to $0.10 per million tokens, a 95% discount compared to standard input.
• The model shows substantial gains in coding, scientific research, and computer-use tasks, often matching or exceeding more expensive competitors.
• Factual accuracy has improved significantly, with a 32% reduction in error rates on difficult prompts compared to GPT-6 Sol.
• Safety alignment is enhanced, showing better transparency about broken tools and stricter adherence to safety constraints.

Sentiment

The community is mixed and skeptical-leaning. There is genuine interest in cheaper, capable models and especially in lower cache costs for agent workflows, but the thread is dominated by doubts about benchmark validity, billing clarity, model naming, and how reliably the promised value appears in real use.

In Agreement

• Lower cached-input pricing is seen as the most practically important part of the announcement because agentic coding often reuses large contexts.
• Some users report that Sol is a strong everyday coding model and a better value than the flagship tier for many real tasks.
• Community examples suggest Sol can be close enough to premium models for practical work, making the price-performance tradeoff attractive.
• Cheaper near-flagship models are viewed as a meaningful step toward making high-quality AI workflows more broadly affordable.

Opposed

• Commenters question whether OpenAI's near-flagship claims hold outside curated benchmarks and want independent real-world evidence.
• Several users criticize the rapid model cadence and branded naming scheme as confusing and difficult to reason about.
• Subscription limits, banked resets, and usage-metering opacity undermine the value story for some paying users.
• Visual and toy benchmark examples still show brittle reasoning, making some commenters skeptical that the model is truly close to flagship quality.
• Some readers interpret the release as a correction for the previous Sol model rather than a straightforward upgrade.