Anthropic Launches Claude Opus 5 and Puts a Dial on Your AI Bill. Here Is What It Means for B2B Teams.

Anthropic released Claude Opus 5 today, and the pitch is aimed squarely at the person signing off on the invoice.

Opus 5 is available immediately across Anthropic's platforms at $5 per million input tokens and $25 per million output tokens, identical pricing to the model it replaces, Opus 4.8. What changed is what you get for that money. Anthropic says Opus 5 "comes close to the frontier intelligence of Claude Fable 5 at half the price," and it becomes the new default model on Claude Max and the strongest model available on Claude Pro (Anthropic, July 24, 2026).

That framing is not an accident. It is a direct response to the single loudest complaint from enterprise AI buyers over the past two months: the bills.

The Token-Burn Hangover

Opus 5 is Anthropic's fourth model release in under two months, following Mythos 5, Fable 5, and Sonnet 5. Fable 5, the most powerful of the group, also became the most debated commercially. Business customers and developers reported the model chewing through enormous token volumes to complete tasks, blowing past allotted budgets and generating bills that were difficult to forecast (Fortune, July 24, 2026).

OpenAI hit the same note when it shipped GPT-5.6 on July 9, emphasizing economical token usage. Two of the three largest model vendors are now competing on restraint. That is a meaningful shift in what the market is buying.

The Effort Dial Is the Actual Product Feature

The mechanism Anthropic is selling is an effort setting, a control that lets customers tell the model how hard to work on a given request, ranging through high, xhigh, and max. Turn it up for intelligence. Turn it down to conserve tokens for faster, cheaper output.

For anyone who has tried to build an AI budget, this is the more important announcement than the benchmark scores. It converts a variable, unpredictable cost into something closer to a lever you can set per workload. Cheap effort for classification and drafting. Max effort for the analysis that goes in front of a client.

Anthropic's benchmark claims are all framed in those terms: performance per dollar, not performance in the abstract (Anthropic, July 24, 2026).

On Frontier-Bench v0.1, Opus 5 surpasses all other models and more than doubles Opus 4.8's performance at a lower cost per task. On CursorBench 3.2 at max effort, it lands within 0.5% of Fable 5's peak score at half the cost per task. On OSWorld 2.0, a computer-use benchmark, it beats Fable 5's best result at just over a third of the cost. On ARC-AGI 3, which tests novel problem-solving, its score is three times the next-best model.

What It Does for Teams That Are Not Engineering Teams

The most relevant data point for B2B marketing, sales, and revenue operations is not a coding benchmark. It is Zapier's AutomationBench, which measures whether a model can complete a business task start to finish.

Opus 5's pass rate is roughly 1.5 times the next-best model at the same cost per task, and even at its lowest effort setting, it passes more tasks than any other model. Zapier CEO Wade Foster described what that looked like in practice: the model took a raw account-health workbook and ran a full churn-prevention sequence end to end, flagging at-risk accounts, alerting the right owner, and summarizing for retention ops. "Previous models didn't pass," Foster said. "Opus 5 hit 100%" (VentureBeat, July 24, 2026).

Box CTO Ben Kus reported an 8% overall improvement over Opus 4.8 on specialized enterprise content, with an 11% gain on data analysis workflows and 17% on due diligence, the work technology, healthcare, and public sector organizations run daily.

Efficiency shows up in customer reports too. One financial-modeling team measured 9 percentage points higher accuracy with a third fewer turns and tool calls, and 60% less time. A legal AI team found Opus 5 held quality while generating 26% fewer tokens on average than Opus 4.8 at max reasoning (Anthropic, July 24, 2026).

Fewer turns, fewer tokens, fewer humans babysitting the loop. That is the enterprise value proposition, stated plainly.

The Behavior Change: It Checks Its Own Work

Anthropic's qualitative claim is that Opus 5 verifies its output and recovers from its own errors without intervention, the difference between a tool and something closer to a colleague.

The examples are specific. Given a drawing of a machine part and no way to view it, the model wrote its own computer vision pipeline to extract the geometry from raw pixels, then rebuilt the part in FreeCAD. No competing model solved it in five attempts. Given a real bug in a popular open-source package manager, it found the root cause and fixed an edge case the community's own patch had missed (Anthropic, July 24, 2026).

Cursor co-founder Sualeh Asif said the model "delivers near Fable 5 intelligence at Opus speed and cost." Lovable co-founder Fabian Hedin pointed at variance rather than peak: 22% better than Opus 4.7 on their hardest agentic coding tasks, and "far less variance run to run." For teams shipping on a schedule, consistency is the number that matters (VentureBeat, July 24, 2026).

The Guardrail Question B2B Buyers Should Ask

Anthropic says Opus 5 is its most aligned model to date, scoring 2.3 on its automated behavioral audit for misaligned behavior, the lowest of its recent models, with the lowest rates of deceptive behavior and the least susceptibility to being tricked into misuse (Anthropic, July 24, 2026).

It is also less likely to refuse legitimate requests. Anthropic expects Opus 5's cyber safety classifiers to intervene roughly 85% less often than Fable 5's. The model can find vulnerabilities in source code, but binary-based vulnerability scanning, penetration testing, and exploit generation remain blocked.

Hugging Face CEO Clem Delangue told Fortune that closed model APIs "flag and refuse a lot of legitimate security work, because analyzing an attack looks a lot like preparing one," and that during an active incident, teams cannot have their tools refusing to examine payloads. Anthropic's answer is automatic fallbacks: flagged requests route to another model rather than returning an error, now available as an API option and on by default in Claude.ai, Claude Code, and Claude Cowork (Fortune, July 24, 2026).

What to Do With This

Three practical takeaways for anyone running AI in a B2B operation.

Re-price your workloads. If you moved work to a cheaper model to control spend, the math just changed. Same price as Opus 4.8, materially better output, and a dial to tune the rest.

Stop measuring cost per token. Measure cost per completed task. A model that finishes in a third of the turns is cheaper at a higher list price, and the vendors are now competing on exactly that metric.

Audit your refusal rate. If your security, research, or compliance teams have been quietly working around a model that will not engage, fallback routing and looser cyber classifiers are worth a fresh evaluation.

Anthropic is reportedly preparing an IPO later this year. Shipping frontier-adjacent capability at half the frontier price, on the same day rivals are fielding questions about agents that compromised a partner's systems, is positioning that happens to line up with what enterprise buyers have been asking for (CNBC, July 24, 2026).

Sources

[Anthropic: "Introducing Claude Opus 5," July 24, 2026](https://www.anthropic.com/news/claude-opus-5)

[Fortune: "Anthropic's Opus 5 targets enterprise AI cost complaints," July 24, 2026](https://fortune.com/2026/07/24/anthropic-claude-opus-5/)

[CNBC: "Anthropic launches Claude Opus 5," July 24, 2026](https://www.cnbc.com/2026/07/24/anthropic-claude-opus-5.html)

[VentureBeat: "Claude Opus 5 benchmarks and enterprise performance," July 24, 2026](https://venturebeat.com/ai/anthropic-claude-opus-5/)