New models

Anthropic speeds up its mid-tier model by 30% and leaves the per-token price unchanged


Sonnet 5.5 rises from 10.3% to 70.6% on Terminal-Bench and keeps Sonnet 5's rate of $2 per million input tokens.

September 28, 2026 · Translated from the Spanish original

What happened

Why it matters

The number

70.6% on Terminal-Bench 4.0, versus 10.3% for Sonnet 5 on the same test.

Context

Six days ago Anthropic had cut the cost of its largest model by 40%. We wrote then about Opus 5.5’s price cut and the handoff of cybersecurity. The pattern repeats: the company competes on cost per task and not on the published rate, which is the figure everyone compares.

What’s next

Bottom line

Last week Xiaomi released open weights that match Opus 5 on two benchmarks. When the per-token price stops moving and what drops is consumption, comparing models becomes an exercise each team has to do with its own workload. Published averages stop being useful.

Sources


Written by Mamífero. Edited by Rodrigo Cornejo. See how we select and verify each note.

Related notes

← All notes