SAN FRANCISCO — Anthropic published its first-week telemetry summary for the Claude Helios rollout Friday afternoon, confirming through aggregate enterprise usage data that code-generation and financial-services verticals have led the model’s adoption, with pharmaceutical and legal verticals following at substantially lower volumes but with higher per-session token consumption.

The telemetry summary, distributed to Anthropic’s enterprise-customer channel and posted to the company’s public model documentation, covers usage from the model’s general-availability launch on May 12 through May 14, the third full operational day. Anthropic chose to publish the summary on Friday afternoon rather than waiting for the natural seven-day window to align with the broader rhythm of the May 8 voluntary disclosure framework, which had been a focus of the company’s communications in recent days.

Code-generation use cases accounted for approximately thirty-eight percent of total Helios inference volume across the three-day window, the company reported, with the highest concentration in enterprise customers in the financial-services and software-development sectors. The average code-generation session involved approximately 4,200 model-generated tokens, slightly below the typical session length for Anthropic’s prior generation of models but with substantially higher reported user-acceptance rates.

Financial-services use cases — defined as enterprise customer accounts in banking, asset management, insurance, and adjacent regulatory sectors — accounted for approximately twenty-three percent of inference volume. Within that vertical, Anthropic identified three principal use patterns: research-document synthesis, regulatory-compliance analysis, and structured-data extraction from unstructured filings. Research-document synthesis sessions averaged the highest token consumption of any reported use pattern, with an average of approximately 11,400 tokens per session.

Pharmaceutical-vertical adoption, while smaller in aggregate inference volume at approximately seven percent, showed the highest average session length of any vertical at approximately 14,200 tokens. The company attributed this to the model’s use in scientific-literature synthesis and in clinical-trial document review, both of which involve substantial document-grounded reasoning over long contexts. Several large pharmaceutical customers had been pre-positioned for Helios access in the period leading up to the May 12 launch.

Legal-vertical adoption accounted for approximately five percent of inference volume in the reporting window. Anthropic noted that the legal vertical’s adoption pattern showed a distinctive concentration in mid-sized law firms rather than in the largest firms, which the company attributed to the longer enterprise-procurement timelines typical at the largest firms. The company said its enterprise sales team expected the legal vertical’s volume to grow substantially over the second and third weeks of the rollout.

A senior Anthropic enterprise-product official, in a background briefing for technology reporters Friday afternoon, said the first-week telemetry was “broadly consistent with the deployment scenarios we had modeled in pre-launch planning” but had surprised the company in two specific areas. The first was a higher-than-expected pharmaceutical-vertical session length, which the official attributed to “the depth of document-grounded reasoning the model is being asked to support.” The second was lower-than-expected gaming-industry adoption, which the company had projected to be a significant early-adopter vertical.

The May 8 voluntary disclosure framework’s sixty-day pre-deployment notification window had been used by Anthropic for the Helios release, with the company having filed its pre-deployment notice on March 8 through the Frontier Model Assurance Council’s interim secretariat. The notice’s substantive content remained confidential under the framework’s standard provisions, but the Council confirmed that the notice had been “reviewed and processed” before the May 12 launch.

A senior Frontier Model Assurance Council official, contacted Friday afternoon, said the Helios launch had been “a positive test of the framework’s operational mechanics” and that the Council had not raised any specific concerns during the pre-deployment review period. The official said the Council was now preparing for the next set of pre-deployment notices, which were expected to arrive from the other signatory companies over the coming weeks.

Industry analysts said the Helios first-week pattern was consistent with the broader market dynamics that had emerged in enterprise AI deployment over the past eighteen months. A senior analyst at a major U.S. technology research firm said the code-generation vertical’s leadership position had become “the structural foundation of enterprise AI deployment,” with each successive generation of frontier model amplifying rather than disrupting that pattern. The analyst said the pharmaceutical and legal vertical patterns reflected a maturation of those industries’ approach to AI-augmented work.

A senior official at one of the four other signatory companies to the May 8 framework, contacted Friday afternoon, said the Helios telemetry release had “set a useful precedent” for the kind of operational disclosure that the company expected to make when its own next-generation model reached general availability. The official did not commit to a specific timeline but said the company was “tracking toward a release in the late-summer window.”

Treasury’s grid-impact disclosure rulemaking, published Thursday morning, had focused industry attention on the infrastructure-side disclosure obligations that would attach to the largest training facilities. The Helios telemetry release, by contrast, addressed the deployment-side question of how frontier models are used after release. Industry observers said the two disclosure frameworks would together constitute the operational backbone of federal-level visibility into the frontier AI ecosystem as it continues to evolve.

Anthropic plans to publish the next Helios telemetry summary at the end of the model’s second operational week, with subsequent summaries expected on a roughly monthly cadence through the model’s general operational life.