The two launches
According to the Last Week in AI channel, OpenAI and Anthropic are bringing more capable AI into their everyday model ranges. In episode 259, recorded on September 30, 2026, Andrey Kurenkov and Jeremie Harris discuss GPT-6.1 Sol and Claude Sonnet 5.5 together. The central theme is the amount of useful work a model can do for its price.
OpenAI introduced GPT-6.1 Sol on September 29. The channel describes it as approaching GPT-6 Astra on important tasks while charging about one-fifth of Astra’s standard token prices. Tokens are the small pieces of text a model processes and produces. This makes the price of repeated requests especially important for agents that perform many steps.
What changes with Sonnet 5.5
Anthropic introduced Sonnet 5.5 on September 28. The channel highlights output that is more than 30% faster than Sonnet 5 and costs up to 30% less for most work. The lower task cost comes from needing fewer tokens, rather than a blanket reduction in the standard token rate.
Both Sol and Sonnet 5.5 are offered at $2 per million input tokens and $10 per million output tokens in their launch materials. Their cached-input rate is $0.10 per million tokens. Caching lets a service reuse previously processed material. The podcast focuses on how these changes make frequent coding and office work more affordable.
Coding results and the wider picture
Kurenkov and Harris also discuss Sonnet’s strong coding results, including tests on which it can exceed Opus 5.5. They point out that the way an agent runs a test and the amount of computation it uses are part of the comparison. The channel distinguishes these measured tasks from the experience of using a model for a whole working day.
In the broader discussion, the channel connects lower inference costs with the economics of running AI services. A cheaper capable model can let a provider serve more requests within the same subscription. The two releases illustrate a shared product direction: bringing stronger capabilities into models intended for frequent use, alongside more expensive models for demanding work.
