Anthropic just made its most capable model a lot cheaper and a lot more private. Opus 5, the latest iteration of the company’s flagship AI, is now available at half the cost of its previous Fable sibling — a pricing move that signals real competitive pressure in the enterprise AI market.
But the price cut isn’t the only headline here. In an unusual move for a frontier model provider, Anthropic confirmed that using Opus 5 does not require data retention. For businesses and developers who’ve been hesitant to send sensitive data through commercial APIs — or who operate under strict compliance regimes — this removes a significant barrier. It’s a direct shot across the bow of competitors whose default terms still involve some level of logging.
The release comes amid a broader industry skirmish over open-weight models. A group of tech leaders just published a letter urging the US government to recognize the strategic value of open-weight AI. Conspicuously absent from the signatories? One name you’d expect to champion openness. The Register notes the letter’s signers include familiar proponents of open ecosystems — but not everyone in Silicon Valley is convinced that giving away model weights is a good idea.
Elsewhere in AI infrastructure, the alliance formations are getting interesting. AMD and Cerebras have teamed up to target Nvidia’s inference dominance, specifically taking aim at Groq’s LPUs. It’s a classic “enemy of my enemy” play. Meanwhile, AMD is also pushing its ROCm software stack forward with what’s being described as “vibe coding” — using natural language prompts to generate optimized GPU kernels, a direct assault on Nvidia’s CUDA moat. Whether developers actually switch toolchains remains the trillion-dollar question.
