Anthropic has released Opus 5, the latest iteration of its popular coding model, and the big news isn't that it's dramatically smarter - it's that it costs about half as much as its fancier sibling, Fable. According to Anthropic's own charts, Opus 5 performs similarly to or slightly ahead of Opus 4.8 and OpenAI's GPT-5.6-Sol on benchmarks like Frontier-Bench and DeepSWE, but it's no Fable-level leap. The main selling point: near-Fable performance at roughly half the price.

Anthropic deliberately held back on cybersecurity training for Opus 5, so it's substantially behind Fable and Mythos on exploiting vulnerabilities - though it's still decent at finding them. That also means it lacks some of Fable's controversial protections, like the 30-day data retention policy for incident review. The model costs $5 per million input tokens and $25 per million output tokens, matching its predecessor but undercutting Fable. However, the Chinese open-weight model Kimi K3 offers similar performance at just $15 per million output tokens, so the competition isn't exactly quaking.

The real story here is cost. As companies like Cursor and Meta build model routers to automatically select cheaper models for simpler tasks, Anthropic needs to keep driving token costs down - or at least offer more performance for the same price - to avoid losing users to smaller or open-weight models that are getting good enough for routine development work.