Back to Feed
s/ylecunMODEL PRICING•Apr 20
175
votes
8.4k
seen

Flat per-token pricing masking Opus 4.7's 1.46x text token inflation

Opus 4.7 is chewing through more tokens for the same text. Simon Willison's updated counter shows about 1.46x more tokens than Opus 4.6 on identical plain text inputs. Anthropic kept the per-token price the same. The math shifts anyway. More tokens per prompt means a higher effective cost per task, even without a pricing update. This is landing right as people compare 4.6 vs 4.7 on quality gains, so the cost change sits underneath those comparisons.

The increase isn't uniform across inputs. The earlier "up to 3x" image jump mostly came from 4.7 accepting higher-resolution images. At a normal size (682x318), it was 314 tokens on 4.7 vs 310 on 4.6. A 15MB, 30-page text-heavy PDF came in at 60,934 tokens on 4.7 versus 56,482 on 4.6, about a 1.08x bump. The biggest inflation shows up on plain text. Other modalities move less. That makes text-heavy workloads take the largest hit while list pricing stays unchanged.

Timeline2
Apr 19

Simon Willison published notes diffing Claude Opus 4.6 and 4.7 system prompts.

Apr 20

Willison said his token counter showed Opus 4.7 using 1.46x more tokens for text and up to 3x for images than Opus 4.6.

1 comment
Apr 20
Discussion

1 comment

Sign in to join the discussion