Within two days at the end of September, Anthropic and OpenAI each released a new mid-tier model, the kind most businesses use for everyday work. If your team or your automations run on either company's tools, the models underneath have probably changed, or soon will. Here's what each release offers and how we'd think about choosing between them.
Claude Sonnet 5.5
Anthropic released Claude Sonnet 5.5 on September 28. Anthropic says it runs more than 30% faster than Sonnet 5 and costs up to 30% less per task, mainly because it needs fewer tokens to finish the same job. Its listed price is $2 per million input tokens and $10 per million output tokens. Anthropic positions it for well-scoped everyday tasks, fixing bugs and producing documents, slides and spreadsheets, and says its larger sibling, Opus 5.5, is still clearly stronger for complex, open-ended work.
GPT-6.1 Sol
OpenAI announced GPT-6.1 Sol at its DevDay event on September 29. OpenAI's API pricing page lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, the same as the earlier GPT-6 Sol, with cached input cut from $0.20 to $0.10 per million tokens. OpenAI describes it as offering near-Astra performance on coding and computer-use work, and its pricing page lists it at one fifth of GPT-6 Astra's $10 and $50 per million token prices.
How they compare in practice
On paper the list prices are identical, so the real cost difference will come from how many tokens each model uses on your work and how often you reuse the same instructions or documents. Both companies' benchmark claims come from their own testing. Neither tells you how the model will handle your quotes, your contracts or your customer emails.
For most small businesses the deciding factor is still where your work already lives. If your team works in Claude or uses Claude-based connectors, Sonnet 5.5 is the natural default. If you're built around ChatGPT, Microsoft tools that use OpenAI models, or OpenAI's API, 6.1 Sol is. Switching providers for a modest gain on a benchmark rarely pays for the disruption.
The caveat
When a vendor changes the default model, unattended automations can behave differently without anyone noticing. Output format, length and how strictly instructions are followed can all shift between versions.
What to do this month
Pick two or three real tasks you run often and run them on the new model alongside your current one. Compare the quality and check the token usage on your bill. If you have automations that run without a person reviewing them, spot-check their output for the next couple of weeks.