Anthropic says Opus 5.5 matches Fable 5.1 “on most tasks” while costing about 40% less to run than Opus 5; Opus 5.5 costs $4/1M input and $20/1M output tokens
Matthias Bastian /The Decoder:
Context & Ripple Effects
Anthropic held Opus 5 at $5 per million input tokens and $25 per million output tokens after emphasizing its coding, reasoning, and prompt-cache-friendly tool use in July 2026. Opus 5.5 lowers list prices to $4/$20 while Anthropic says it reduces running cost by about 40%.
The comparison is sharper because Anthropic priced Fable 5.1 at $10/$50 earlier in September, despite a 75% cut to cache-read pricing in its Fable 5.1 pricing update. Opus 5.5 positions near-Fable task performance as the lower-cost option rather than merely a cheaper incremental upgrade.
First-order effects
- Anthropic customers can shift suitable workloads from Opus 5 to Opus 5.5 at lower token prices while retaining Anthropic's claimed Fable 5.1-level performance on most tasks.
- Fable 5.1's $10/$50 headline token rate faces a more direct internal benchmark after Opus 5.5's $4/$20 pricing and Anthropic's 40% lower-running-cost claim.
Second-order effects
- Anthropic's enterprise buyers and agent builders gain a stronger incentive to measure model selection by completed-task cost, including retries and cache use, rather than by nominal model tier.
- Anthropic must substantiate the gap between token pricing and running-cost claims for workloads that previously relied on Opus 5's coding and tool-use improvements.
Third-order effects
- If comparable task performance continues to move down the price curve, frontier-model competition will center less on list token rates and more on effective cost per reliable agent task.
- Safety controls become part of the economic comparison: Anthropic's stated enhanced safeguards can affect the usable cost of deployment alongside raw inference pricing.
The trend: Frontier AI vendors are compressing the price premium for high-end performance, shifting buyer attention toward effective inference cost and verified task outcomes.