Anthropic now offers four Claude models, each optimized for different types of work, and picking the wrong one can quietly double your monthly bill or cap the quality your customers see. The lineup includes Haiku 4.5 for simple, high-volume tasks; Sonnet 5 as the default workhorse; Opus 4.8 for long-running autonomous work; and Fable 5 for the hardest problems. The key insight from teams running these models in production is that the model name matters far less than matching the task to the right price and capability level.

What Are the Four Claude Models and How Much Do They Cost?

Anthropic's four models sit at deliberate points on the price and capability curve. Output tokens are where the real cost accumulates, since models generate far more tokens than they read on most jobs. The pricing spread is stark: Fable costs fifty dollars per million output tokens, while Haiku costs just five dollars per million output tokens, making Fable ten times more expensive to run at scale.

• Haiku 4.5: Costs one dollar per million input tokens and five dollars per million output tokens, with a 200,000-token context window. Best for high-volume, simple, repetitive work like document classification, field extraction, and tagging.
• Sonnet 5: Costs three dollars per million input tokens and fifteen dollars per million output tokens (with an introductory rate of two dollars in and ten dollars out until August 31, 2026). Carries a full one-million-token context window and serves as the sensible default for most production work.
• Opus 4.8: Costs five dollars per million input tokens and twenty-five dollars per million output tokens. Designed for long-running autonomous agents and complex tasks that need to maintain reliability over many steps without human intervention.
• Fable 5: Costs ten dollars per million input tokens and fifty dollars per million output tokens. The most capable model with reasoning always enabled, reserved for the single hardest problem in a project where Sonnet and Opus both fall short.

Which Claude Model Should You Actually Use for Your Workflow?

The mistake most teams make is picking a model by reputation and then applying it everywhere. A cheap model on the right job beats an expensive one on the wrong job every time, on both cost and reliability. For Australian businesses and others paying in US dollars, the routing decision should be based on task shape, not model prestige.

Haiku 4.5 shines when a task is simple, repetitive, and needs to happen fast at scale. One finance and accounting automation pipeline uses Haiku to read thousands of invoices a day and extract supplier, amount, and tax line by line. At one dollar in and five dollars out per million tokens, the running cost barely registers against the labor it replaces. The one caution is the context window: Haiku tops out at 200,000 tokens, a fifth of the other models, so it is the wrong tool if a single task needs to hold a very large document in memory.

Sonnet 5 is the model most teams should start with. It delivers reasoning close to Opus on coding and agentic work, yet costs significantly less. The introductory pricing through August 31, 2026, makes the value even more compelling. Most automations shipped by production teams run on Sonnet by default, and the interesting question is only where to step up or down from it.

Sonnet 5 also represents a meaningful step forward in agentic capabilities and tool use. The model can now dispatch multiple tool calls simultaneously rather than sequentially, significantly reducing latency in multi-tool workflows. It is also better at diagnosing when a tool call fails and choosing an appropriate recovery strategy rather than getting stuck in retry loops. On coding benchmarks, Sonnet 5 significantly closes the gap with Opus 4: on SWE-bench Verified, which tests practical ability to fix real issues in real GitHub repositories, Sonnet 5 jumped from 49.3% to 62.4%, compared to Opus 4.8 at 71.2%.

Opus 4.8 earns its keep on work that runs a long time without a human in the loop. Multi-step research agents, complex refactoring across a large codebase, and planning tasks that branch and backtrack are where the durability premium justifies the cost. For professional services firms running overnight analysis, that reliability over a long horizon is worth the extra expense. The rule with Opus is to prototype the agent on Sonnet first, prove the logic, then move only the long-running production path to Opus if Sonnet drifts over many steps.

Fable 5 is the most capable model Anthropic ships, with reasoning always enabled, and it is the most expensive to operate. It also consumes shared weekly usage limits faster than the rest of the range, so it is not a model you leave running on everything. Keep it for the single hardest problem in a project, the one where Sonnet and Opus both fall short and the outcome justifies the cost. If you find Fable running across a whole pipeline, that is usually a sign the pipeline was never routed properly in the first place.

How to Route Tasks to the Right Claude Model

• Start with Sonnet 5: Make it your default model for most production work. The model delivers strong reasoning on coding and agentic tasks at a reasonable price, and the introductory rate through August 31, 2026, offers exceptional value.
• Step down to Haiku for simple, repetitive work: If a task is high-volume, simple, and does not require a large context window, Haiku will cut your costs dramatically while maintaining quality on straightforward jobs like classification and field extraction.
• Step up to Opus for long-running agents: If your task runs for hours without human intervention and small early mistakes compound over time, prototype on Sonnet first, then move to Opus only if Sonnet drifts over many steps.
• Reserve Fable for the hardest problem: Use Fable only when Sonnet and Opus both fall short and the outcome justifies the ten-fold cost increase. Treat it as a scalpel, not a default.
• Monitor token usage in production: Before committing to any model at production volume, run your token estimate through an AI ROI calculator so the monthly figure is real rather than a guess.

Sonnet 5 also now comes as the default model across all Claude.ai tiers, including the free plan, replacing Sonnet 3.7 as the standard. This decision represents a significant philosophical stance by Anthropic: frontier capability should be broadly accessible, not gatekept entirely behind paywalls. For developers building autonomous AI workflows, Sonnet 5's improvements to tool use and agentic state management address the most common failure modes in production AI agents.

The broader lesson from teams running real client workloads across all four models is consistent: the model name matters far less than the shape of the task. Choosing the right model by task, not by reputation, is how you avoid quietly doubling your monthly bill or capping the quality of work your customers see.