The new model arrived on July 24 with a price tag that did not move: $5 per million input tokens and $25 per million output tokens, identical to the Opus 4.8 it replaces. That single fact tells the story of Claude Opus 5. In a market where every frontier lab races to claim the smartest model, Anthropic chose to sell something else — control over how much intelligence a customer actually pays for.
Opus 5 ships with adjustable effort settings, from low to max, that let users trade capability for token savings depending on the task. Routine work can run at lower effort, cutting reasoning tokens and cost. Complex problems can still command the model’s full attention. The same weights, the same model ID, and bills that vary by orders of magnitude depending on how they are called.
Ars Technica’s assessment cut to the point: this does not look like a capability jump, it looks like an efficiency optimization. Anthropic positions Opus 5 as coming close to the frontier intelligence of its pricier Claude Fable 5 — launched in June at $10 and $50 per million tokens — at half the price. On at least one independent benchmark index, Opus 5 at maximum effort scores slightly above Fable 5, according to figures cited by model trackers.
The release marks a deliberate divergence from the industry’s raw-performance arms race. OpenAI, Google and Meta have spent the past year pushing flagship models with ever-larger benchmarks and premium prices. Anthropic’s move reframes the competition: instead of asking which model is smartest, it asks which model delivers the most capability per dollar — and lets the customer set the dial.
Enterprise buyers have been asking exactly that question. AI bills have become a boardroom issue as companies moved from pilots to production workloads, and finance teams have pushed back on costs that scale with usage. Anthropic’s answer is a model that makes the trade-off between quality and cost explicit, a pitch aimed squarely at chief financial officers who approved the pilots.
The model also consolidates Anthropic’s product ladder. Opus 5 becomes the default on Claude Max and the strongest model available on Claude Pro, while Fable 5 is reserved for the longest, most autonomous jobs. On financial-modeling tasks, the company said Opus 5 averaged nine percentage points higher accuracy with a third fewer turns and tool calls and 60 percent less time compared with Opus 4.8.
Pricing strategy has become the quiet battleground of the AI industry. Over roughly a year, Opus-tier pricing has fallen threefold, and each major lab has cut API rates to win developer loyalty. Anthropic’s gamble is that flexibility will matter more than headline benchmarks as workloads industrialize.
The model’s efficiency focus reflects a deeper shift in customer behavior. Early adopters bought AI capability as a novelty; production users buy it as an input with a unit cost. For those buyers, a model that can be tuned to the difficulty of each task is not a luxury — it is the difference between a service that makes money and one that burns it.
Analysts said the launch responds to a specific threat: competitors undercutting on price. OpenAI’s flagship pricing and Google’s aggressive bundling have put pressure on every independent lab. Anthropic’s response is differentiation through control — a product that cannot be compared on a single benchmark because it is configured per use.
The model’s adaptive thinking, enabled by default, bills reasoning tokens as output, a detail that changes real invoices. Two teams on identical pricing can run up very different bills depending on how aggressively they tune effort levels. That is precisely the point: Anthropic is selling a spend dial, and the company’s own documentation now reads like a budgeting guide.
The reception will test a theory about the AI market. If enterprises reward the efficiency pitch with adoption, other labs will follow with similar controls, and the industry’s center of gravity will shift from raw capability to unit economics. If buyers keep gravitating to the smartest available model, Anthropic’s differentiation will look like a side bet.
The company is not abandoning the frontier. Fable 5 remains its top-tier offering, and a restricted model called Mythos 5 sits above both, available only in limited settings. But the product that most businesses will actually run — the default on the highest-volume plans — is now the one engineered to cost less.
Developers get the model through the Claude API as claude-opus-5, with a fast mode that runs about 2.5 times standard speed at double the base rate. Prompt caching floors dropped to 512 tokens, making cached calls cheaper for workloads with stable prefixes. Every detail of the launch is tuned to one variable: the bill.
The industry will know within a quarter whether the strategy worked. Usage data and enterprise adoption will settle the question faster than any benchmark, according to analysts who track API traffic. In the meantime, Anthropic has done something unusual in the AI race — it slowed down, on purpose, to sell efficiency in a market that assumed speed was the only product.
