Choosing your Claude model: the simple version

Claude now has four distinct models that match specific jobs instead of one size that tries to do everything. Fable, Opus, Sonnet and Haiku are the names and each has a clear role, price and context window. Fable and Opus share a 1,000,000 token context and are designed for heavy multi-step work and deep reasoning respectively. Sonnet also has a 1,000,000 token context and sits as the versatile daily default with an introductory price through 31 August 2026 ( personally use this as default). Haiku drops to a 200,000 token context with a 64,000 reply length and is built for high-volume quick tasks. Concrete figures matter because billing now reads like a taxi meter. Prices are listed per million tokens: Fable at $10 in and $50 out, Opus at $5 in and $25 out, Sonnet at $2 in and $10 out through 31 August 2026 before moving to $3 in and $15 out, and Haiku at $1 in and $5 out. Plans matter too because Fable requires a paid plan while Opus needs Pro and above and Sonnet and Haiku are available on Free and above.

Why should this model split change how you work with Claude? The answer is practical: choice saves money and reduces friction when results matter. You can stop overpaying for routine tasks and stop trusting lighter models with legal level reasoning when precision counts. You will notice cost differences fast when you run high-volume jobs because Haiku costs about a fifth of Sonnet for simple outputs and Sonnet remains the best low-friction option for most office work. Opus is the step up for documents that need cross-checking or careful logic and Fable is the reserve for projects that are long, multi-step or require handing a full project over to the model. You’ll also get better results more consistently because you pick for the task instead of defaulting to one model. That becomes the operational skill that saves money and preserves quality and it scales whether you’re a solo consultant or a small team of five.

Who will notice the most immediate gains from this clarity? Solo founders, consultants, executive assistants and small agency teams will benefit first because they run a wide range of tasks and tight budgets. You as a business owner will save on repetitive work by using Haiku for tagging, short replies and high-volume lookups. You as an EA or VA will find Sonnet handles most daily workflows like drafting emails, research, and light automation and you can keep costs predictable on the Free plan (but limited). You as an analyst or technical consultant will reach for Opus when reasoning must hold up under scrutiny, and you’ll reserve Fable for large, multi-step projects that you'd rather hand off than micromanage.

Is adopting the model ladder worth the learning curve? Yes when you treat it like a three-rule escalation and no if you ignore the pricing and context differences. The ladder is: Sonnet first, Haiku for volume, Opus for hard reasoning, and Fable for very large complex projects. You’ll get immediate ROI because Sonnet handles the majority of tasks and is available on the Free plan. Haiku saves money on repetitive volume work so your monthly token spend drops. Opus is costlier but worth it when the alternative is manual review by a human at a higher hourly rate. Fable is expensive but designed for handing off whole projects so it can replace external contractor fees on long efforts. You should test this over two weeks with three representative tasks: one routine task on Haiku, one multi-step workflow on Sonnet, and one complex reasoning task on Opus. Track time saved and token spend, and adjust the routing rule. That small experiment reveals whether the new model split returns value to your specific workflows.

Get the most bang for buck by starting on Sonnet for everyday tasks and routing high-volume simple work to Haiku to save money. You should step up to Opus only when Sonnet fails a clear accuracy or reasoning test and reserve Fable for long, multi-step projects you want the model to manage end to end. One person will need to be assigned (quick, don't make eye contact) to own the routing rule and a simple spend dashboard that tracks in and out token usage. That governance step is cheap but decisive because it prevents wasted spend and preserves output quality. The approach works for basically anyone from solo operators to small agencies and avoids the trap of overpaying for capability you don’t require.