UPGRADE DECISION — AUGUST 31 DEADLINE
● Capability: Sonnet 5 is strictly better — adaptive thinking, better agentic performance, 128K output
● Current cost (through Aug 31): Sonnet 5 at $2/M is cheaper than Sonnet 4.6 at $3/M
● Post-deadline cost (from Sept 1): Sonnet 5 at $3/M = same sticker as 4.6 — but new tokenizer adds 10-35% more tokens
● Tokenizer uplift (Sonnet 5): 10-35% more tokens for same text vs Sonnet 4.6
● New restriction: Sonnet 5 returns 400 error if you set temperature/top_p/top_k to non-default — check your API calls
● Verdict: Switch now and front-load volume before August 31
Per the official Sonnet 5 release notes, Sonnet 5 is a strict performance improvement over Sonnet 4.6 on every capability dimension. It is also the default model for Free and Pro users as of June 30, 2026. The cost question is more nuanced. Through August 31, Sonnet 5 at $2/M is actively cheaper than Sonnet 4.6 was at $3/M — and capability is better. From September 1, the sticker price matches ($3/M for both), but the new Sonnet 5 tokenizer maps the same text to up to 35% more tokens. For code-heavy or non-English workloads, effective cost from September 1 could be higher than Sonnet 4.6 at its $3/M standard price despite identical per-token rates. One migration caution: Sonnet 5 returns a 400 error if you set temperature, top_p, or top_k to non-default values. If your existing Sonnet 4.6 calls set any of these parameters, remove them before switching to Sonnet 5 or your calls will fail.
Switch to Sonnet 5 now. It is better in every capability dimension and currently cheaper. Front-load volume before August 31.
Last updated August 1, 2026. Related: Sonnet 5 deadline action plan →