Claude Sonnet 5 Ends the Opus Monopoly — Pricing Sunsets Too

Claude Sonnet 5 launched June 30, 2026 — and the model tier that used to be the budget compromise is now Anthropic’s default agent: it plans, drives browsers and terminals, and runs autonomously at a level that previously required the flagship Opus line. The pricing structure does the sunsetting in two directions at once: intro rates of $2/$10 per million tokens expire August 31 (rising to standard $3/$15), and the old Sonnet generation’s era of “cheap mid-tier” ended with it.

Key facts

• The release: Sonnet 5, June 30, 2026 — “our most agentic Sonnet yet” — with a 1M-token context window, 128K max output, and a new tokenizer.
• The positioning: near-Opus 4.8 performance at a fraction of flagship cost; default model for Free and Pro plans, in Claude Code, and on the API as claude-sonnet-5.
• The pricing cliff: $2/$10 intro through August 31, then $3/$15 — still undercutting Opus 4.8 ($5/$25) and GPT-5.5, but a 50% jump for anyone who built on intro rates.
• The API change-up: effort levels (low–max) are the cost/latency control, adaptive thinking is always on, and temperature/top_p/top_k requests now return 400 errors — migration code required.
• The legacy sunset: Sonnet 4.5 reaches end-of-support September 29, 2026, and Opus 4.1 hit EOL August 5 — Anthropic’s deprecation conveyor is now clockwork.

Why it matters

• The mid-tier is the battleground. Sonnet pricing is the price of running agents at volume — and $2/$10 is the shot across the bow at the entire agentic cost layer. When the intro ends, every agent company built on Sonnet 5 gets re-rated.
• API behavior changed with the model. Anthropic removed sampling knobs in favor of effort levels — a philosophical statement: the model decides how much to think. Every consumer of the API must rework request payloads.
• The sunset clock forces decisions. Locking in workloads now vs. paying +50% in September is the strategic spreadsheet question of the month.

What to watch

• September 1, 2026 — whether standard pricing holds or the $2/$10 sticks (hear the launch bet followed by a “we heard you”).
• Agent default wars — Sonnet 5 as default model while Claude Cowork pushes always-on autonomy: prompt-injection surface grows with adoption.
• The deprecation cadence — Sonnet 4.5/Opus 4.1 retirements while Opus 5 (July 24) absorbs flagship users.

Official source

• Anthropic: Introducing Claude Sonnet 5
• TechCrunch: Anthropic launches Claude Sonnet 5 as a cheaper way to run agents
• Model deprecations: Anthropic Claude platform docs

Updated August 10, 2026.