Anthropic cuts Claude Sonnet 5 pricing ~60% to $2/$10 per million tokens

Anthropic slashed Claude Sonnet 5 pricing to $2 per million input tokens and $10 per million output tokens, effective through August 31, a cut of roughly 60% from prior rates. The move comes as Anthropic reports an approximately $47B annualized revenue run rate and faces an intensifying price war across the frontier — GPT-5.6's Luna tier, Grok 4.5, Meta's Muse Spark 1.1, and DeepSeek V4 all pitched aggressive economics the same week.

The developer reaction was notably positive. On Hacker News and Reddit, the price cut drew broad praise, with developers calling agentic capability at $2/$10 per million tokens 'a genuine value shift' that enables mid-market teams to migrate off pricier flagships. That contrasts with the more skeptical reception GPT-5.6's government-gated preview received on the same platforms.

Strategically, the timing is defensive as much as offensive. With Meta explicitly threatening a coding price war and DeepSeek and Alibaba's Qwen undercutting on the open-source side, Anthropic needs to keep Sonnet — its high-volume workhorse — competitive on cost while reserving Opus and Fable for premium tiers. The 'through August 31' framing suggests a promotional window rather than a permanent repricing, giving Anthropic room to reassess. It also arrives amid a rough week for Anthropic's China standing (the Claude Code backdoor allegations and Alibaba ban) — making a visible pro-developer gesture strategically useful. The broader read: even the labs with the strongest coding reputations now compete primarily on dollars-per-task, confirming that the industry mood this week is cost, not capability, as the differentiator.