Simon Willison

Highlight

The economics of frontier AI are shifting, as high costs for top-tier models like Anthropic’s Fable are forcing developers to move away from blindly using the most powerful model toward cost-effective routing and optimization. This trend is backed by newly reported revenue figures and model-spend indexes showing that cheaper or older models are dominating actual adoption.

Posts

Anthropic’s best AI model struggles to attract users as cheaper tools thrive · Source Simon unpacks a Financial Times story detailing Anthropic’s massive annualized revenue leap to $65 billion in July (up from $47 billion in May) alongside OpenAI’s surge to over $40 billion. Despite these soaring figures, Anthropic’s premium Fable 5 model is struggling with low adoption due to its high cost. Simon highlights data from the Ramp AI Index showing Fable 5 capturing only 8.0% of Anthropic model spend, confirming that developers are increasingly opting for cheaper, “good enough” alternatives like older Opus versions.

Quoting Drew Breunig · Source Simon highlights Drew Breunig’s piece on how the premium pricing of Anthropic’s Fable has marked “the end of the free lunch” in AI-assisted programming. Previously, developers could ignore optimizing context strategies or coding harnesses, knowing a cheaper, more powerful model would soon arrive to paper over inefficiencies. Now, with Fable’s high cost barrier, engineers must actively design smart routing systems to send tasks to cheaper, capable models like Opus, GPT 5.6, K3, or GLM.

Project Pulse

This day’s posts reveal a clear focus on the shifting economics of AI development, specifically the transition from “lazy” upgrading to deliberate, cost-aware model routing. Both pieces highlight how the high cost of frontier models like Fable is pushing developers to optimize their context strategies and build smarter harnesses for cheaper, capable models.

📈 We could do some web research on how the Ramp AI Index has evolved since July 2026 to see if Fable 5 or other frontier models have managed to capture more developer spend.