Anthropic launched Claude Opus 5 on July 24, 2026, pricing it at $5 per million input tokens and $25 per million output tokens — exactly half the cost of Claude Fable 5 — while matching or beating Fable 5 on two leading benchmarks, Frontier-Bench v0.1 and OSWorld 2.0, at maximum reasoning effort. The model carries a 1 million token context window and introduces five per-request reasoning effort levels — low, medium, high, xhigh, and max — that let API callers trade compute depth against cost on every call.
What Claude Opus 5 Costs
Claude Opus 5 is priced at $5.00 per million input tokens and $25.00 per million output tokens in standard mode. A fast mode is available at approximately $10.00 per million input tokens and $50.00 per million output tokens — twice the cost, with roughly 2.5 times the throughput speed. Fable 5 is priced at $10.00 per million input tokens and $50.00 per million output tokens with no fast-mode tier. Opus 5 delivers near-Fable-5 intelligence at Fable 5’s fast-mode price in standard mode, or at that cost with faster throughput in fast mode.
How the Reasoning Effort Settings Work
Claude Opus 5 gives API callers five per-request effort levels: low, medium, high, xhigh, and max. Low effort is faster and cheaper for routine tasks such as text classification or template filling. High and max effort allow the model to reason longer on complex tasks such as multi-step data analysis or code generation. Anthropic designed the setting as a per-request parameter, so a single application can run customer-support queries at low effort and financial-modeling queries at max effort without switching models. The feature is Anthropic’s direct response to tiered reasoning APIs such as OpenAI’s o-series.
How Opus 5 Benchmarks Against Fable 5
At maximum reasoning effort, Claude Opus 5 scores within 0.5% of Fable 5’s peak score on Frontier-Bench v0.1 and matches or beats Fable 5 on OSWorld 2.0, according to benchmark data published at launch. Artificial Analysis places Opus 5 at an Intelligence Index score of 61 and ranks it first overall on its model leaderboard. Secondary sources including DigitalApplied and BenchLM confirm the Frontier-Bench and OSWorld results; exact numerical scores have not been published on a primary benchmark page. Fable 5 remains the absolute performance ceiling; Opus 5 approaches it at half the per-token cost.
What This Means for Business API Users
Claude Opus 5 is Anthropic’s fourth model launch in under two months, following Mythos 5 (coding and agentic tasks), Fable 5 (flagship tier), and Sonnet 5 (mid-range). Each model fills a distinct price-performance band. For teams evaluating the best AI tools for business, Opus 5 changes the cost calculus: a pipeline that previously required Fable 5 to meet accuracy thresholds can now run at half the per-token cost in standard mode, or at the same cost as Fable 5 with 2.5× faster throughput in fast mode.
Teams building AI agents for multi-step reasoning can set effort at max for complex reasoning steps and drop to low for retrieval and formatting steps — reducing per-session spend without model changes. Anthropic stated at launch: “Claude Opus 5 reaches roughly Claude Fable 5–level intelligence at half the price.” OpenAI’s GPT-5.6 Sol Max holds the #2 position on the Artificial Analysis leaderboard; Opus 5 at $5/$25 makes the cost-per-intelligence comparison directly relevant for enterprise API budget decisions.
For Context
Anthropic released Claude Fable 5, the flagship model that Opus 5 now benchmarks against, in June 2026. Fable 5 was Anthropic’s first publicly available Mythos-class model and set the performance ceiling the company has been building its model ladder below. Separately, Anthropic’s $1.2T valuation — established at its latest funding round — provides context for the pace of the company’s release cadence and infrastructure investment.
Our Take
Opus 5’s effort slider gives business teams a cost dial, not just a capability dial. The real story is that Anthropic is building a pricing structure that resembles cloud compute more than AI subscriptions — pay for the compute depth you need, per request, at five granularity levels. Enterprise API buyers who were priced into Fable 5 purely for accuracy now have a half-price alternative that matches it on leading benchmarks. OpenAI’s fixed-rate model is now visibly under pressure on the cost-per-intelligence axis.
