Anthropic has launched Opus 5, an updated AI model aimed particularly at coding and software development. The article argues that the release is chiefly significant for improving token efficiency and value rather than delivering a major new leap in capability, reflecting growing pressure on AI providers to reduce costs as cheaper and open-weight alternatives improve.

Benchmark results place Opus 5 around the level of Anthropic’s Fable model on coding tasks and generally ahead of Opus 4.8 and OpenAI’s GPT-5.6-Sol, although gains appear incremental. It costs $5 per million input tokens and $25 per million output tokens, while it was deliberately not trained to the same cybersecurity level as Fable or Mythos; competitors including Kimi K3 and automated “model routers” are intensifying the focus on using cheaper models for routine work.