Opus 5 dropped overnight
Same price as previous Opus generations. More capable. In some cases better than Fable at half the price.
Superior to GPT-5.6 Sol in most cases. These are internal benchmarks, so there are reasons to be sceptical.
As we saw with VW’s Dieselgate, it is often tempting to ‘optimise’ (aka cheat) against benchmarks. There are also questions over whether benchmarks are added to the training data of models.
OpenAI dropped SWE-bench Verified as a primary reporting target in February 2025 because the model is “increasingly contaminated.”
I’ll put Opus 5 through its paces in coming days. I don’t expect much discernible difference as I’m already satisfied with Opus 4.8. Same cost. More capable.
Update for Claude Cowork
Added /usage and /cost to Cowork tasks: an inline card shows your plan limits and the session’s usage without sending anything to the model. (21 July)
Type /cost and hit return to see how costly a particular session was.
Improved Voice Mode
Anthropic also just shipped a meaningful update to Claude’s voice mode, and the headline feature is model choice - users can now pick between Opus, Sonnet and Haiku rather than being locked to Haiku by default.
That matters because the original voice mode was fast but shallow. Haiku is built for speed, not depth, and complex tasks showed it. The new default behaviour is smarter: voice mode reads whichever model you last used in text chat and picks its fastest variant. So if you’ve been working in Sonnet, voice inherits that context.
The use cases Anthropic is pointing to - rehearsing a client pitch, getting feedback on how you communicate, talking through product research - signal a deliberate push upmarket. This is voice mode positioned as a thinking tool, not just a hands-free convenience.
The bigger deal is app integration. Claude’s voice mode can now reach into Gmail, Google Calendar, Slack, Canva and Notion, meaning you can verbally ask it to reschedule a meeting, draft an email or spin up a Notion doc. OpenAI’s updated voice mode, for all its conversational polish, still can’t do this. It’s stuck in chat-only mode. That’s a genuine functional gap, not a marketing one.
Sources:
Anthropic: Claude Opus 5 Anthropic updates Claude voice mode with more capable models