October 11, 2026
Grok Bot: search and plans in the bot, with Opus 5.5 for architecture at twice the price of Sonnet 5.5
Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Sonnet 5.5 costs half as much: $2 and $10 (Anthropic pricing page, Oct 12, 2026). The scheme splits work by task. Architecture, complex code and frontend go to Opus, daily debugging and tests to Sonnet.
Min Choi
@minchoi
This is literally my new workflow: Real-time research and planning → Grok Bot Work assistant → Grok Bot Architecture → Opus 5.5 Day-to-day code and debugging → Sonnet 5.5 Writing and running tests → Sonnet 5.5 Complex code → Opus 5.5 Frontend and design → Opus 5.5 Bookmark it.
· 63K views
Grok Bot covers search and planning. Architecture, complex code and frontend go to Opus 5.5, daily debugging and tests to Sonnet 5.5. The expensive model goes where a mistake costs more than the tokens.
The price gap. Opus 5.5 costs twice as much as Sonnet 5.5 for both input and output. Cached context costs 0.05 of the base input price on both models, so a long debugging session gets cheaper if you don't restart it from scratch.
Access. Grok Bot is open to Starlink and SpaceX customers and to xAI enterprise customers. According to a 36Kr article, the bot has no API fee. Cursor's page on Grok calls the bot a harness the model was trained for and does not say which models the bot uses. The wwwhatsnew review says that from Oct 6 the bot routes complex tasks to Opus 5.5. Those without the bot can still use the scheme itself: you can split models by task in any tool with a model picker.
The hard facts end here: Anthropic's prices can be verified, but the models behind Grok Bot are not officially named.
Source