Claude Opus 5.5 Takes the Lead on Intelligence Index as Anthropic Pushes Run Cost Down
Anthropic introduced Claude Opus 5.5 on September 22, 2026. Artificial Analysis lists the Adaptive Reasoning, Max Effort, Default Fallback configuration with an Artificial Analysis Intelligence Index score of 58, ranked No. 1 of 206 models. Artificial Analysis
The listing identifies Anthropic as the creator and gives September 22, 2026 as the release date. It describes a model with text and image input, text output, and reasoning capability, with a 1M-token context window. Artificial Analysis Anthropic platform documentation dates Claude Opus 5.5 system prompts to September 22, 2026. Anthropic
Pricing is where the release gets specific. Artificial Analysis lists $4.00 per 1M input tokens and $20.00 per 1M output tokens, with a 95% cache discount. Artificial Analysis That compares with $5 per million input tokens and $25 per million output tokens for Claude Opus 5, unchanged from Claude Opus 4.8. Anthropic
Anthropic states that Claude Opus 5.5 costs 40% less to run than Opus 5. List prices alone imply a 20% reduction. The difference matters for anyone modeling inference budgets, because run cost reflects throughput, caching behavior, and reasoning length as well as token price.
Artificial Analysis puts the weighted average cost per Intelligence Index task for the tested configuration at $5.98. Artificial Analysis The evaluation generated 260M output tokens and cost $8708.20 to run. Artificial Analysis Those figures are large. They also make the tradeoff explicit: top-of-table accuracy at max effort is expensive to measure, and expensive to reproduce in production if every request uses the same settings.
Positioning is also explicit. Anthropic states Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work. The prior generation was framed more cautiously. Anthropic described Claude Opus 5 as a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5. Anthropic It separately described Claude Opus 5 as its strongest Opus model yet for powering long-running, highly capable agents with improvements in coding and professional work.
The lineage is short and fast. Anthropic described Claude Opus 4.5 as its flagship model with state-of-the-art coding and agentic performance at lower pricing. Anthropic Anthropic announced Claude Opus 5 as available on all platforms at the same pricing as Opus 4.8. Anthropic Its system card described Claude Opus 5 as its latest large language model and an upgrade to Claude Opus 4.8. Elsewhere in the portfolio, Anthropic states Claude Sonnet 5's introductory pricing of $2 per million input tokens and $10 per million output tokens has since been made permanent. Anthropic
External comparison adds one data point. Sacks said GLM-5.2 is "just a tick below Anthropic's Opus 4.8 and right up there with OpenAI's GPT 5.5." Reuters
The broader context here is cost per unit of capability, not rank alone. A No. 1 score at max-effort reasoning tells buyers what is possible when token budget is unconstrained. It does not tell them what the same model costs at default effort, with aggressive caching, or on long-context retrieval where input tokens dominate. Anthropic's emphasis on lower run cost and near-Fable performance suggests where it wants that conversation to go.
In my view, the naming does useful work. Adaptive Reasoning, Max Effort, Default Fallback signals three separate decisions that operators now have to make on every frontier deployment: how much thinking to allow, how hard to push it, and what to do when constraints bind. Watching my own children adopt these tools, the pattern is familiar. They use the smartest setting once, then leave the cheaper default on. Enterprise buyers tend to follow the same path, once the invoice arrives. That makes the $5.98 per task figure as important as the 58. It frames what frontier reasoning actually costs to use, not just to benchmark, and what becomes possible if that cost keeps falling while capability holds.
