Choosing a Claude model has become its own small purchasing decision. Anthropic’s new guide gives us something more practical than “use the biggest one”: a starting point, cheaper options for routine work, and a warning about turning the thinking dial too high.
A clearer guide to Claude’s models
Anthropic’s Claude model guide recommends Opus 5.5 at medium effort for most coding and knowledge work. It is already the default in Claude Code across Pro, Max, Ultra, Team and Enterprise plans, as well as the Anthropic API.
For everyday documents, slides and spreadsheets, Anthropic recommends Sonnet 5.5. Complex financial models or documents assembled from a large pile of material move into Opus territory. Haiku 5.5 is the company’s pick for high-volume sorting, tagging and extracting information; Fable 5.2 is for the hardest technical projects and long unattended runs.
That makes this useful even if you never write code. A recurring drafting task and a complicated financial analysis needn’t get the same model budget. Nobody needs another settings hobby.
For API users, the price differences are concrete: Sonnet costs $2 per million input tokens and $10 per million output tokens; Opus costs $4 and $20; Fable costs $8 and $40. Those are API list prices, separate from subscription pricing.
Anthropic also recommends adjusting effort before switching models. Higher effort gives the model more room to think and check its work, but adds waiting time and cost. On its SWE-bench Pro tests, Opus at extra-high effort scored about 1.4 points above high effort while costing 2.5 times as much.
Otter adds languages in stages
Otter’s multilingual expansion adds 27 languages, bringing its planned total to 31. Thirteen are available now, including Mandarin Chinese, Korean, German, Portuguese and Vietnamese.
The bigger change is still upcoming: Otter says it will automatically detect language changes within a meeting and continue transcribing without someone adjusting settings. That is relevant if your customer calls or team meetings regularly move between languages, rather than staying in one throughout.
Existing Otter users can check the newly available languages now, but should wait for the multilingual feature before counting on automatic switching. Another 14 languages are coming soon, including Arabic, Hindi, Thai and Hebrew.
Instagram filters from a prompt
Instagram’s Effect Builder is available now on the web for US residents over 18. It generates custom filters for Stories from a written description, with optional reference images.
For anyone already making Stories for work, this offers a way to create a reusable visual effect. Describe it, upload a photo to preview the result, then request changes through additional prompts.
Publishing has a gate: submitted effects must pass a safety review. Instagram says approved effects become available for anyone to use in Stories, and each use links back to the creator’s profile.
Also
- Medium’s TK writing app. Medium says its new web editor is free with a Medium account and lets writers publish notes explaining their AI use.
- A beep located. Simon Willison says his Claude Code-built system located a beeping carbon monoxide detector within about 40 centimeters using several listening devices.
- Azure document extraction. Microsoft’s agentic mode is in preview and uses additional processing to inspect evidence, calculate and validate document results.
- Robot practice. NVIDIA says GR00T 2 achieved 71% average success across 50 tasks with one demonstration per task.
Pass it on
