Anthropic’s new guide to choosing a Claude model, published on 9 November and written by Addy Osmani, lists Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens, and Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100K tokens. By our arithmetic, Fable 5.2 costs 80 times Haiku 5.5 per token at that tier.
The guide says to “start most work on Opus 5.5” at its default medium effort, priced at $4 and $20, and puts Sonnet 5.5 at $2 and $10. Each step down the ladder halves the price until Haiku, by our arithmetic. Haiku 5.5, the guide says, “is also good to use with subagents that work on a larger model’s side tasks”.
Business Pill · Tokens Explained
An evergreen Business Engineer explainer on tokens. It teaches the general idea only and does not discuss Anthropic’s model guide.
The key insight: As we read it, Anthropic’s guide turns model choice into a margin decision for builders: by our arithmetic Fable 5.2 costs 80 times Haiku 5.5 per token on short prompts, and the guide’s own run shows about 41% less per task from running cheap first and re-running failures.
All four generally available models have a 1M-token context window and 128K tokens of maximum output, according to the guide’s side-by-side table. Haiku 5.5 is the only one whose price changes with prompt length: over 100K tokens it bills at $0.50 and $2.50, five times its short-prompt rate.
Cache reads cost $0.20 per million tokens on both Fable 5.2 and Opus 5.5, $0.10 on Sonnet 5.5 and $0.01 on Haiku 5.5 up to 100K, the guide says. It adds that “on long runs most of your input bills as cache reads”, so on a cache-heavy run the Fable-Opus gap comes down to uncached input and output.
Claude API list prices per million output tokens in Anthropic’s 9 November 2026 model guide: Fable 5.2 $40, Opus 5.5 $20, Sonnet 5.5 $10 and Haiku 5.5 $0.50 for prompts up to 100K tokens. The 80x ratio is our arithmetic.Cost per task for Opus 5.5 on a SWE-bench Pro subset, as reported in Anthropic’s guide: $0.29 with every task at high effort (95.3% passed) against about $0.17 running at low and re-running failures at high (about 97%). The 41% saving is our arithmetic. Source: Anthropic, claude.dev, 9 November 2026.
Our new guide breaks down when to use Fable 5.2, Opus 5.5, Sonnet or Haiku. It also covers when to adjust effort instead of switching models, and how to save cost by putting subagents on smaller models.
Anthropic’s @ClaudeDevs account posted the guide at 21:03 UTC on 9 November, saying it breaks down when to use Fable 5.2, Opus 5.5, Sonnet or Haiku, when to adjust effort instead of switching models, and how to save cost by putting subagents on smaller models. The prices and results in this piece come from the guide it links.
Where the Guide Says to Save
The guide’s first lever is effort, not model. “Before you move models, try adjusting your effort level”, it says. On a SWE-bench Pro subset, it reports, Opus 5.5 at low effort with failures re-run at high passed about 97% of tasks for about $0.17 each, against 95.3% for $0.29 with every task at high. That is about 41% less per task, by our arithmetic.
The second lever is subagents. “You pay for its tokens at the rates of the model it runs on”, the guide says, and it recommends keeping small models on work where a mistake is cheap to spot, such as finding files, running tests and reading logs, while code edits and judgment calls stay on the main model.
For work bigger than one context window, the guide cites an internal defect-search benchmark over about 22 million tokens of source code: an Opus 5.5 lead over Sonnet 5.5 workers, up to 25 at a time, “cost about a third less than Opus 5.5 alone at its default effort” and scored about the same.
Where the Guide Says to Spend
The guide says to move up to Fable 5.2 “for the hardest technical projects in coding, finance and the sciences”. On FrontierCode 2.0, it reports, Fable 5.2 at high scored 45.3% against 45.5% for Opus 5.5 at max, for roughly half the cost per task.
It also warns that the hardest tasks carry most of the spend: “On one 20-problem search benchmark, two problems carried 43% of the spend”. Its advice is to price the hardest tasks in a set as well as the typical ones.
The Structural Read
As we read the guide, the cheapest lever it names is effort rather than a model switch: it says to try adjusting effort before moving models, and that past high, one more effort level can cost as much as a bigger model.
As we read the guide, the subagent advice puts the price gap to work inside one product: it says you pay for a subagent’s tokens at the rates of the model it runs on, and recommends Haiku 5.5 or Sonnet 5.5 for subagents that search files, read logs or summarize.
Anthropic, claude.dev, 9 November 2026
“Keep the expensive model’s context for the decisions, and let cheaper subagents do the verbose reading.”
Three Implications
FOR BUILDERS ON THE CLAUDE API The guide lists Fable 5.2 at $8/$40, Opus 5.5 at $4/$20, Sonnet 5.5 at $2/$10 and Haiku 5.5 at $0.10/$0.50 per million tokens for prompts up to 100K.
FOR CLAUDE CODE USERS The guide says to set model: haiku or model: sonnet in a subagent’s definition, and that the built-in Explore subagent now runs on the session’s model instead of always on Haiku.
FOR TEAMS MEASURING SPEND The guide says two problems carried 43% of the spend on one 20-problem benchmark, and recommends pricing the hardest tasks as well as typical ones.
The framework records: “That ratio is the gross-margin equation of the agentic era.”
As we read it, Anthropic’s guide is a playbook for that ratio: it prices four models from $0.10 to $8 per million input tokens and tells builders which work to route to which. The framework predates the guide and does not discuss it.
What Is Not Established
The cost results are Anthropic’s own runs and benchmarks, not independent tests. The guide does not give a dollar figure for its subagent savings in Claude Code, and the 80 times ratio applies only to Haiku 5.5 prompts up to 100K tokens; above that, Fable 5.2 costs 16 times Haiku 5.5 per token, by our arithmetic.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Anthropic’s guide prices its four models on a ladder from $0.10 to $8 per million input tokens and tells builders to tune effort first, put subagents on Haiku or Sonnet and keep Fable 5.2 for the hardest work; by our arithmetic, Fable 5.2 costs 80 times Haiku 5.5 per token on short prompts.
A note on sourcing. We read Anthropic’s guide on claude.dev of 9 November 2026 in full, and the @ClaudeDevs post that announced it. Prices are the Claude API list prices the guide gives; the ratios and the 41% saving are our arithmetic. The benchmark and cost results are Anthropic’s own runs, which we did not reproduce.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
