Anthropic launches Claude Fable 5.1 and says it's up to 45 percent cheaper for agentic work

Anthropic just made its most capable model significantly cheaper to run, and that matters more than the performance bump. Claude Fable 5.1 costs about 25 percent less than its predecessor in typical use, but the real story is the up to 45 percent savings on complex agentic tasks, according to The Verge. That discount comes from reduced pricing on cached data, the information the model has already processed and stored in memory.

This is Anthropic directly addressing the cost problem that has kept many teams from deploying AI agents at scale. Agentic workflows, where AI systems plan, execute multi-step tasks, and iterate on their own work, rack up token costs fast because they loop through the same context repeatedly. Cheaper caching means those workflows become economically viable for more use cases.

The company also launched Mythos 5.1 alongside Fable, and both models reportedly dial back the overzealous safety guardrails that frustrated developers. Early testers like Every CEO Dan Shipper are calling Fable 5.1 the strongest coding model they've used, praising its speed, token efficiency, and more natural communication style. That combination of performance, cost, and usability could shift which model developers reach for when building production systems.

The timing aligns with broader industry momentum toward agentic AI. OpenAI, Google, and others are racing to make their models better at autonomous work, but cost has been the quiet bottleneck. If Anthropic can deliver agent-grade performance at a price that scales, it changes the math on what kinds of automation projects make business sense.

What this means for you: if you've been holding off on building AI workflows because of cost concerns, Fable 5.1's caching discount makes repeated, context-heavy tasks much more affordable. Try this prompt to prototype an agentic workflow: "You are a research agent. I'll give you a topic, and you should outline a plan to gather information, then execute each step, summarizing findings as you go. Topic: [your research question]. Keep track of what you've already learned to avoid repeating searches." The caching will kick in as the agent references prior context, cutting your costs while it works through the task.