Weekly ranking, Technology

OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

minimaxir.com, Technology

A data-backed case that open-weight competition, not subsidy, drives LLM price cuts, though some figures go unsourced.

What it adds

It grounds the commoditization argument in concrete numbers: OpenRouter provider counts, Vercel cost-per-token data and cache-read arithmetic. It also rejects the Uber-style loss-leader theory, arguing that inference has low barriers to entry.

Claim by claim

• 9 the author's own
• 1 an AI would say
• 2 known elsewhere
• 4 common knowledge

Its strongest original claims

• In agentic coding, cache reads are 96-99% of token spend, so a 75% cache read cut yields roughly 70-75% lower cost.
• Anthropic stated Opus 5.5 requires less compute to serve than Opus 5 and priced it at $4/$20/$0.20.
• The Luna price cut was likely aimed at competing with small open-weight models like DeepSeek V4 Flash.
• Providers hosting DeepSeek Flash on OpenRouter grew from 13 to 29.
• Vercel data shows cost-per-token fell to 43.3% of April 2026 levels while aggregate spend rose.

Scored by Expulp Verdict for what the piece adds beyond what is already known and published. The full verdict shows every claim with its reasons. Leaderboard, what Verdict gets right and where it is weak.