Tariff Tracker: Anthropic's Opus 5 Pushes Token Efficiency, Not Capability — Cost Battle Reshapes AI Spending

1. What happened: Anthropic launched Opus 5, a new version of its popular coding-focused AI model. The model offers roughly the same performance as the pricier Fable model but at about half the cost per token — $5 per million input tokens and $25 per million output tokens. The release is a cost efficiency play, not a breakthrough in capability, as benchmarks show only iterative gains over prior versions. The move comes amid rising competition from open-weight models like Kimi K3 at $15 per million output tokens.

2. Who: Anthropic rolled out the model. The company chose not to train Opus 5 on cutting-edge cybersecurity tasks, leaving it well behind the more capable Mythos 5 and Fable models in that area. The broader AI market includes companies like Cursor and Meta, which are building model routers that automatically select cheaper models for simpler tasks. This pricing pressure affects all frontier model providers.

3. Tickers / sectors: No direct public company tickers are mentioned in the input facts. The story impacts the broader AI and technology sector, including companies that develop or use large language models — but no specific equity tickers appear in the article. The trend toward cheaper, open-weight alternatives could reduce revenue growth for frontier model providers like Anthropic (private) and OpenAI (private).

4. Winners / losers: Winners include companies building model routers (e.g., Cursor, Meta) and providers of open-weight models (e.g., Kimi K3). Losers could be frontier model companies that cannot lower token costs to match smaller, open models for routine coding tasks. Software engineering teams and enterprises win by having more cost-effective options for everyday development.

5. What to watch: Watch for further price cuts from Anthropic, OpenAI, and Google on their frontier models. If model router adoption accelerates, usage of cheaper models may grow, potentially slowing growth for higher-priced options. The trend toward open-weight models could also attract regulatory attention as AI capabilities spread more broadly.