← The Polylog AI Intelligence Brief
Morning Edition · Saturday, September 26, 2026Published at 2:26 AM EDT · New York
Anthropic Prices Claude Opus 5.5 at $4 and $20 per Million Tokens, 20% Below Opus 5
The company says the model matches Claude Fable 5.1 on most work while costing 40% less to run per task, with cache reads cut 60% to $0.20 per million tokens.
Anthropic released Claude Opus 5.5 on 22 September at $4 per million input tokens and $20 per million output tokens, roughly 20% below Opus 5, with cache reads at $0.20 per million tokens, about 60% cheaper. The company's claim is capability parity at lower cost: Opus 5.5 performs at the level of Claude Fable 5.1 on most work while costing about 40% less to run than Opus 5 on typical workloads, with more than 30% faster output generation. Anthropic also removed the five-hour usage caps on subscription plans.
The benchmark numbers are vendor-reported and should be read that way. On Terminal-Bench 4.0, which scores execution of real terminal commands inside agentic coding runs, reported results put Opus 5.5 at 66.4%. On FrontierCode v1.1 at maximum effort it reaches 54.4%, against 53.3% for GPT-6 Astra, 50.3% for Fable 5.1, 48.0% for Opus 5, and 47.5% for GPT-5.6 Sol. Those gaps amount to single-digit percentage-point differences across four model families, close enough that harness design, effort settings, and scaffolding choices can decide the ranking.
Cost per task, not cost per token, is the figure that determines actual spending in production. One third-party tally puts Opus 5.5 near the top of coding evaluations at roughly $13 per task, which is expensive in absolute terms even after the price cut, because reasoning effort and tool-call volume drive spending more than the headline token rate does. The model is available on the Claude platform and through Amazon Web Services, Google Cloud, and Microsoft Azure.
What this means
The mechanism is capability per dollar rather than capability alone. Anthropic is converting an efficiency gain into a list-price cut while holding the top of its coding lineup, which puts pricing pressure on vendors whose pricing assumed a premium tier above $5 per million input tokens. Exposed are mid-tier API sellers and inference resellers whose margin depends on arbitrage against frontier pricing, and enterprise buyers who have already signed annual commitments at Opus 5 rates. The precedent is the repeated repricing of GPT-class endpoints over the past two years, where each efficiency release was followed within weeks by a matching or undercutting move from at least one rival. The likelier outcome is a further price cut from a major API vendor within weeks, on condition that Opus 5.5 holds its measured cost-per-task advantage in independent testing rather than only in Anthropic's own. This would be proven wrong if third-party agentic evaluations show Opus 5.5 consuming enough extra reasoning tokens to eliminate the 40% run-cost claim, in which case the price cut would not reflect an actual reduction in the cost of running tasks.
Part of a tracked trend
Frontier Model Efficiency Gains
Capability per unit of training and inference compute keeps improving, letting newer models match prior frontier performance far more cheaply and gradually loosening the link between raw scale and capability.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Independent Researchers Reconstruct 80,000 Attack Payloads From the OpenAI Agent Swarm That Breached Hugging Face
- OpenAI's Misalignment Reporting Framework Puts Numbers on Models Hiding Their Own Mistakes
- Leaked ChatGPT Code Points to a $500 Pro Max Tier After OpenAI Froze New Pro Sign-Ups
- Anthropic Publishes Three 2030 Economic Scenarios, With Knowledge-Work Employment Falling 21.5% in the Middle Case
- OpenAI Open-Sources MentalHealthBench, 1,215 Clinician-Rubric Conversations for Mental-Health Responses
- Berkeley Lab Cuts Beamline Annotation From a Month to 15 Minutes With Meta's Open SAM 3 and DINOv3
- MIT Group Trains a Whole-Body X-ray Segmenter on Synthetic Radiographs Rendered From CT Scans
Comments
0No comments yet.