Anthropic's Claude Opus 4.7 Costs Significantly More Per Request Despite Flat Per-Token Pricing

Anthropic has released Claude Opus 4.7 with a new tokenizer that breaks text into up to 47 percent more tokens than the previous Opus 4.6 model, resulting in significantly higher costs per request for users. Despite maintaining identical per-token pricing between the two versions, each request to Opus 4.7 costs substantially more due to the increased token count for the same text input. Early measurements reveal the impact of this tokenizer shift particularly affects Claude Code users, who face elevated operational expenses. The pricing structure creates a situation where flat per-token rates mask the real cost increase driven by the tokenization methodology change.

Opus 4.7's Higher Effective Costs Despite Flat Per-Token Pricing

• ▪Anthropic's Opus 4.7 costs significantly more than Opus 4.6 despite flat pricing.
• ▪Anthropic's Opus 4.7 matches its predecessor's per-token price.
• ▪Each request to Anthropic's Opus 4.7 costs significantly more than its predecessor despite matching per-token pricing.

Impact of the New Tokenizer on Token Counts

• ▪Early measurements show the impact of Anthropic's Opus 4.7 tokenizer shift for Claude Code users.
• ▪Anthropic's Opus 4.7 uses a new tokenizer that breaks the same text into up to 47 percent more tokens than its predecessor.

Perspective of Claude Code users

• ▪Claude Code users experience higher per-request costs when using Anthropic's Opus 4.7 compared to Opus 4.6.
• ▪The tokenizer change in Anthropic's Opus 4.7 directly impacts the operational expenses of Claude Code users.

1 source