Anthropic is running an introductory rate on Claude Sonnet 5 — $2/M input, $2/M cached input, $4/M output — through August 31, 2026. The standard rate of $3/$15 takes effect after. At the intro price, Sonnet 5 is the cheapest mid-tier frontier model with Anthropic’s agentic coding strength.
The Rate Card
| Period | Input $/M | Cache Read $/M | Output $/M |
|---|---|---|---|
| Intro (thru 8/31/26) | $2.00 | $2.00 | $4.00 |
| Standard | $3.00 | $3.00 | $15.00 |
Why Sonnet 5 Matters
Sonnet 5 is the practical daily-driver of the Claude 5 family — close to Opus 5 on agentic benchmarks at a fraction of the cost, and the strongest mid-tier coding model available. With a 200K context window and top-tier tool calling, it is the workhorse for production agent loops.
The Timing Play
If you are evaluating ANthropic’s mid-tier, the discount window is the time to run evals and lock in workloads before the $3/$15 rate applies. Teams that standardize on Sonnet 5 now can budget the standard rate from September while validating at 2x-3x lower output cost this month.
How It Compares
At intro pricing, Sonnet 5’s output rate ($4) undercuts GPT-5.6 Terra ($12) and Gemini 3.5 Pro ($8) significantly. Even at standard $15, it remains competitive on coding quality per dollar. The Claude family positioning is in our AI Model Comparison 2026.
Bottom Line
August is the cheapest month to test Claude Sonnet 5. Sign up, benchmark your actual tasks, and decide before August 31.
