The bill for running frontier artificial intelligence is starting to arrive, and the company most exposed to it is doing something unusual: rather than racing to the top of a benchmark leaderboard, Anthropic is racing to the bottom of a price-per-token spreadsheet. On Friday, the San Francisco AI lab launched Claude Opus 5, a model it says delivers nearly all of the reasoning power of its flagship Claude Fable 5 at roughly half the operating cost, and is positioning the new release as the workhorse enterprises should reach for every day.
Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, the same rate as its predecessor Opus 4.8. It is available immediately across Anthropic’s full platform stack and becomes the default model on Claude Max and the strongest option available on Claude Pro. For a vendor whose enterprise API business has become the financial center of gravity, holding the price flat while improving capability is the move that matters, and Anthropic is making sure customers notice.
The Economics Argument
Anthropic is not pretending Opus 5 is its smartest model. That crown still belongs to Fable 5, and rival systems retain advantages in select domains. Instead, the company is advancing a more pragmatic thesis: that the most economically important AI work sits in a middle band of difficulty, where near-frontier intelligence delivered cheaply beats frontier intelligence delivered expensively. “Opus 5 as your daily driver, the model you hand complex work to and review when it’s done,” an Anthropic spokesperson told VentureBeat, framing the lineup as Opus 5 for daily use, Fable 5 for the longest autonomous jobs, Sonnet 5 for high-volume work, and Haiku 4.5 for subagents and instant answers.
The numbers on the launch slides reflect that posture. On Frontier-Bench v0.1, an agentic terminal coding benchmark, Opus 5 scored 43.3 percent, more than double Opus 4.8’s 18.7 percent and well ahead of Fable 5’s 33.7 percent, at a lower per-task cost. On ARC-AGI 3, an evaluation of novel problem solving, Anthropic reports Opus 5 scored three times as high as the next best model. On OSWorld 2.0, a computer-use benchmark, the model surpassed Fable 5’s best score at just over a third of the cost. The benchmarks also expose the model’s ceiling: Anthropic concedes Opus 5 trails a competitor called Mythos 5 on cybersecurity and biology research, and an OpenAI-family model still leads on one agentic coding evaluation.
What The Benchmarks Miss
Asked directly where Opus 5 falls short of Fable 5, the spokesperson offered a candid answer that may end up defining the next phase of model competition. “The evals where Opus 5 wins are bounded tasks with a specific outcome, which is where it’s strongest. What those evals don’t measure is duration,” the spokesperson said. “One way to put it: Opus 5 is the best tool for the jobs benchmarks can see, and Fable 5 is what you reach for when the job outruns the benchmark.” Fable 5, the spokesperson added, is built for “the longest, most autonomous jobs, where the model has to stay coherent across many connected steps over hours or days with dense source material,” and the company is urging customers to “run both on a representative workload, one bounded task and one long-horizon job.”
That distinction between bounded tasks and long-horizon autonomy may quietly become the most important axis of differentiation in the AI market in 2026, as traditional benchmarks saturate and the hardest remaining problems shift toward multi-day agentic work.
Early Customers Lean Into The Token Math
Enterprise early adopters are publicly backing the efficiency case. Harvey, the legal AI company, said Opus 5 matched Opus 4.8’s maximum-reasoning performance “while generating 26% fewer tokens on average,” according to Niko Grupen, head of applied research. Richard Pham of Fundamental Research Lab reported that on difficult financial-modeling tasks, the model averaged nine percentage points higher accuracy “while using roughly one-third fewer turns and tool calls and 60% less time.” Wade Foster, CEO of Zapier, said Opus 5 topped his company’s AutomationBench leaderboard “without spending more tokens than prior Claude models” and ran a full churn-prevention workflow end-to-end. “Previous models didn’t pass; Opus 5 hit 100%,” he said. Scott Wu, CEO of Cognition, the company behind the Devin coding agent, said on FrontierCode 1.1, “Claude Opus 5 approaches Fable-level performance at half the cost,” with notable strength in debugging and root-cause analysis.
The launch arrives as Anthropic’s API and enterprise business has become the dominant slice of its revenue. According to a February 2026 analysis by Contrary Research, Claude held roughly 40 percent of the enterprise large language model market by usage in late 2025, and Claude Code alone had reached about $1 billion in annualized revenue.
Anthropic’s Behavioral Story Beneath The Numbers
Beneath the benchmark claims, Anthropic is selling something more qualitative: that Opus 5 verifies its own work and iterates until it succeeds. The company shared several illustrative episodes. In one Frontier-Bench task, asked to reconstruct a machine part as a 3D CAD model from a drawing it had no way to view, the model wrote its own computer vision pipeline to extract geometry from raw pixels, and did so repeatedly while no competing model solved the task in five attempts. In another, given a real bug in a popular open-source package manager, the model found the root cause and fixed an edge case the community’s own patch had missed, while a competing model patched only the symptom and declared victory.
Customers reported similar behavior in production. Cristian Rivera, a staff software engineer at Stripe, said he gave the model “a chief-of-staff role over my dev environments” for a weekend: “it built its own monitor, drove each box, and pulled me in only for the judgment calls.” An engineer at a trading firm used Opus 5 to build a market data feed for a new exchange in a single session, and, finding no live feed to validate against, watched the model construct its own test harness to verify its parsing code.
The takeaway for enterprise buyers is sharper than any benchmark chart: as inference costs become a board-level line item and the most valuable AI work happens in the bounded middle of the difficulty curve, the model that wins the daily contract is the one that delivers the most usable intelligence per dollar. With Opus 5 priced the same as the model it replaces while clearing its predecessor on coding and knowledge-work evaluations, Anthropic is betting that math is the only benchmark that ultimately matters.
Anthropic is leaning directly into that shift, treating price-per-token as a product surface rather than a billing detail, and the Claude Opus 5 launch reads less like a capability arms-race moment and more like the opening move in a slower, more deliberate campaign to make frontier reasoning feel routine. By bundling aggressive subscription economics with enterprise controls like SSO, audit logs, and data-residency commitments, Anthropic is signaling that the next phase of competition will be decided inside procurement spreadsheets and CFO memos, not just on leaderboards. The bet is simple: if every knowledge worker eventually has a default AI assistant, the vendor that wins the contract will be the one whose usage feels costless at scale. Anthropic is pricing accordingly, and rivals will be forced to answer.