Anthropic Locks Sonnet 5 at $2/$10 as Price War Intensifies

The frontier model price war is no longer speculation — it is now confirmed policy at both top labs. On August 11, Anthropic permanently locked Claude Sonnet 5 at its June 30 introductory rate of $2 per million input tokens and $10 per million output, scrapping the $3/$15 increase scheduled for September 1. The move comes twelve days after OpenAI slashed GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% (OpenAI, 2026).

Sonnet 5 pricing locked permanently

When Anthropic launched Claude Sonnet 5 on June 30, the $2/$10 pricing was explicitly labeled introductory — a promotional rate set to expire August 31 and rise to $3/$15 on September 1. On August 11, Anthropic quietly made that introductory rate permanent, updating its pricing page without a dedicated blog post.

Anthropic did not publicly explain the decision and did not respond to press inquiries (The Stack, 2026). The lock follows OpenAI's July 30 efficiency-driven cuts across the GPT-5.6 family (OpenAI, 2026):

Sonnet 5 now undercuts GPT-5.6 Terra on output tokens ($10 vs $12) while matching it on input ($2). For agentic coding workloads — where output token counts dominate — that $2 gap per million tokens compounds quickly.

Why it matters

Three dynamics shifted this week.

The introductory-rate bluff is called. When two top labs cut prices within twelve days of each other, it is not a promotion — it is a new price floor. Frontier model pricing is now a competitive lever, not a fixed premium.

Sonnet 5 becomes the clear mid-tier default. At $2/$10, Sonnet 5 is priced below GPT-5.6 Terra on output and matches it on input. For teams running agentic coding pipelines, Sonnet 5 at the locked rate is the strongest value in the closed-model mid-tier — and arguably the full Anthropic lineup.

Enterprise budgets get certainty. Teams that standardized on Sonnet 5 expecting the August 31 price hike no longer need a contingency plan. The September increase was going to sting migrated workloads. That uncertainty is gone.

The counterintuitive result: per-token costs are in freefall, but total agent spend may stay flat or rise. Agentic workloads consume more tokens than ever — each task can spawn dozens of sub-agent calls — so cheaper per-token pricing finances more ambitious automation, not lower bills.

What changes for you

If you are on Claude Sonnet 5: the introductory pricing risk is off the table. You standardized on $2/$10 — and now that rate holds. No migration, no budget recalibration, no September contingency needed.

If you are on GPT-5.6 Terra: compare. Sonnet 5 now costs the same on input and less on output ($10 vs $12). For agentic workloads where output tokens dominate, the gap is meaningful. Run your own evals — Sonnet 5's Terminal-Bench 2.1 score of 80.4% (Anthropic, 2026) is competitive with Terra's independently-evaluated CLI-coding performance, and the price advantage may tip the balance.

If you are evaluating both: the competitive landscape just shifted. Mid-tier pricing is converging around $2/$10–$12, and both labs are signaling that further cuts are on the table. Do not budget against a single vendor's list price — build for price discovery.

FAQ

Is the $2/$10 rate really permanent?
Anthropic has not specified a new expiration date, and the pricing page now lists the rate without a promotional qualifier. "Permanent" in frontier AI has a short shelf life — but for the foreseeable planning horizon, this is the rate.

Will OpenAI respond with another cut?
Almost certainly. OpenAI tied its July 30 cuts to infrastructure efficiency gains and explicitly framed lower prices as mission-aligned (OpenAI, 2026). If Sonnet 5 at $2/$10 pulls meaningful workload from Terra, expect another round. Neither lab can afford to be undercut for long.

Does this affect other Anthropic models?
This pricing lock applies only to Claude Sonnet 5. Anthropic has not announced changes to Opus 4.8 or Claude Haiku pricing.