Between 1 September and 7 October 2026, Anthropic released four new Claude models: Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. If you open the model menu in Claude and wonder which one to pick, you are not alone. The short answer is that most people should leave everyday work on Opus 5.5, use Sonnet or Haiku when speed or cost matters, and save Fable for the hardest problems. The longer answer is worth a few minutes, because choosing well saves you usage limits on a subscription and real money on the API.
The four models in one sentence each
Claude Fable 5.1 is Anthropic's top model, which it describes as the world's most advanced for coding and knowledge work. It is also the most expensive. A sibling called Claude Mythos 5.1 is the same model with looser safeguards for vetted security and life-science professionals, and most people will never see it.
Claude Opus 5.5, released on 22 September, is the first model of the new 5.5 family. Anthropic says it performs at the level of Fable 5.1 on most work while costing 40% less to run than Opus 5. That combination is why we think it is the right default for most people.
Claude Sonnet 5.5, released on 28 September, is the faster, cheaper complement to Opus. Anthropic says it runs more than 30% faster than Sonnet 5 and costs up to 30% less for most work.
Claude Haiku 5.5, released on 7 October, is the small model. Anthropic calls it the cheapest, fastest and most capable small model it has released, built for high-volume work like summaries, sorting and sub-tasks inside larger agents. It costs around 75% less to run than Haiku 4.5.
Standard prices from Anthropic's model pages, October 2026. Output costs five times the input price for every model: $50, $20, $10 and $0.50 per million tokens. Haiku's price applies to prompts up to 100,000 tokens.
What the price gap means if you only use the app
If you use Claude through a Pro, Max or Team plan, you don't pay per token. But your plan's usage limits work in a similar way: heavier models use up your allowance faster. Picking a lighter model for simple jobs is the easiest way to avoid hitting a limit in the middle of the afternoon. When Opus 5.5 launched, Anthropic also raised the five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, so the default model now goes further than it did in the summer.
A simple rule for picking
Think about how much it would cost you if the answer were slightly wrong. If the answer is low stakes and you'll read it anyway, such as a summary of a long email thread, a rewrite of a paragraph or sorting feedback into themes, use Sonnet 5.5 or Haiku 5.5. They are fast enough that you'll notice the difference, and good enough that you won't notice the quality gap.
If the task needs judgement across a lot of material, such as reviewing a contract, planning a launch, analysing a spreadsheet or writing something you'll send to a client, use Opus 5.5. This is where most knowledge work sits, and it is why we would make Opus your default.
If the task is genuinely hard and you have already seen Opus struggle with it, such as a tricky bug across a large codebase, a long chain of reasoning or an agent working on its own for a long time, switch to Fable 5.1 for that task and then switch back. Paying the Fable price for a routine email is wasted money or wasted limits.
For builders: speed modes and the new price floor
If you use the API, two changes stand out. Opus 5.5 has a fast mode that Anthropic says is up to 2.5 times faster, at $8 per million input tokens and $40 per million output tokens, double the standard rate. That is useful for chat products where people are waiting for a reply. And Haiku 5.5 at $0.10 per million input tokens makes it realistic to run a model over every support ticket, review or log line you have. Anthropic also halved Sonnet 5.5's cache read price to $0.10 per million tokens, which it says makes Sonnet about 20% cheaper on most agentic work.
A common pattern is to let Opus 5.5 plan a task and hand the repetitive parts to Haiku 5.5 running as sub-agents. You get Opus-level judgement where it matters and Haiku-level cost for the bulk of the work. If you are on a Max or Team plan, the monthly API credits included since October are a free way to try this.
What we would do this week
Open Claude's model menu and check what your default is. Then run one task you do every week on two models, side by side, and notice whether the difference matters to you. Most people find that a lighter model is fine for more of their work than they expected, and that Opus is noticeably better on the tasks they care about most. That one comparison tells you more than any benchmark. If you want to see how Claude compares with ChatGPT and Gemini for the same jobs, our Compare page does exactly that, and our piece on OpenAI's GPT-6 models explains the equivalent choice in ChatGPT.