GPT-6.1 Sol: near-Astra work at ~1/5 the price

OpenAI launched GPT-6.1 Sol as an upgrade to GPT-6 Sol: near–GPT-6 Astra scores on agentic coding (DeepSWE), professional PDFs (GDP.pdf), multi-step business workflows (AutomationBench), and computer use (OSWorld), at about one-fifth of Astra’s standard input/output token prices. API list prices are $2 / $0.10 cached / $10 per million input / cached input / output tokens; model id gpt-6.1-sol. Available now in ChatGPT Work and Codex for Plus through Edu; not yet in Chat. Ultrafast version coming. The system card addendum treats it as Critical for cybersecurity (same safeguards stack as Astra) and High for bio/chem; OpenAI reports alignment closer to Astra than to GPT-6 Sol on several agentic checks. Vendor benches — verify on your tasks. HN discussion (~807 points, ~745 comments): one camp celebrates the cache cut and “Astra-class work at Sol economics”; another says Opus 5.5 still wins real coding and that GPT-6 Sol was a regression, so they will wait on 6.1 vibes. A third argues DeepSeek / open models already beat $200 plans on bang-for-buck. Soft consensus: price is real news; capability claims need personal evals, not only benches. Sorted take: if you run OpenAI agents for coding, PDF-heavy finance/legal work, or computer use, make 6.1 Sol your default Work/Codex model this week and keep Opus or Astra only for the hardest slices — measure cache hit rate and end-to-end task cost, not just the list price.