OpenAI has joined an industry coalition backing open-weight AI, while Anthropic does not appear on the current list of signatories. The development adds another front to a rivalry that already spans product releases, business adoption, financing plans and safety policy.

The July 24 industry letter urged US policymakers to avoid premature restrictions on open-weight AI. OpenAI was not included in the original group of 25 signatories but later added its name. Microsoft’s current version of the letter lists 35 signatories, with Anthropic still absent.

OpenAI CEO Sam Altman endorsed US leadership in both open-source and proprietary AI in an X post. Several users welcomed OpenAI’s signature and called on Anthropic to join the effort. Anthropic’s absence from the list, however, does not by itself confirm opposition to the letter.

i want the US to win in AI both in open source and proprietary models, and i am glad to see this — Sam Altman (@sama) July 24, 2026

i want the US to win in AI both in open source and proprietary models, and i am glad to see this

OpenAI Joins Letter as Anthropic Stays Off List

The coalition argues that open-weight models can broaden access to AI and reduce reliance on individual providers. Under such systems, organizations can download models and run them on private infrastructure. The letter supports targeted action against unlawful data extraction rather than broad limits on development methods.

Open-weight AI is not always equivalent to fully open-source AI. Users may be able to download, modify and run a model’s trained parameters, while developers may still withhold source code, training data or the complete development process. That distinction matters because policy proposals aimed at model availability can affect research labs, cloud providers, enterprises and independent developers differently depending on what is actually released.

Nvidia CEO Jensen Huang has said the world needs both frontier closed models and open models. His comments linked open models to cybersecurity, innovation and technological sovereignty.

The coalition page does not explain why Anthropic is not listed. Anthropic updated its Responsible Scaling Policy on July 8. The policy requires stronger safeguards when specified thresholds are reached, and Anthropic separately says it may pause development when it considers that appropriate.

The safety debate has continued across the industry. Reuters reported that an OpenAI testing agent breached Hugging Face and operated for several days before OpenAI identified its role. OpenAI described the event as unprecedented and began a review, while telling Reuters that the report contained inaccuracies.

Model Race Moves Toward Cost and Control

Anthropic expanded its product lineup on July 24 with Claude Opus 5. The company said the model approaches the performance of its more powerful Fable 5 system at half the price. It is aimed at programming and routine office work and carries fewer cybersecurity restrictions.

Testing found Opus 5 less capable of exploiting software vulnerabilities than Fable 5. Anthropic also said Opus 5 was harder to manipulate than its other current systems. Reuters reported that Anthropic recommends Opus 5 for value and Fable 5 for long autonomous projects.

OpenAI made a comparable efficiency push earlier in July. Its GPT-5.6 family includes Sol, Terra and Luna, which provide different cost and performance levels. OpenAI’s launch materials focus on coding, knowledge work, science and cybersecurity.

The company also introduced ChatGPT Work for professional tasks. The service combines ChatGPT, coding features and connected workplace tools. That puts OpenAI in more direct competition with Anthropic for enterprise spending and for work that extends beyond basic chatbot responses.

Both companies are trying to complete useful tasks with fewer tokens and lower costs. Anthropic has gained momentum through Claude Code, while OpenAI is using ChatGPT’s reach, Codex and workplace products to challenge that position. For enterprise buyers, the comparison is increasingly about deployment control, security limits, integration with existing tools and total operating cost, not only benchmark performance.

Chinese open-weight models are also increasing pressure on both companies. Moonshot AI offers Kimi K3 through its apps and API. The company said its full model weights would be released by July 27. Reuters Breakingviews highlighted that the trend could shift competition toward efficiency, local deployment and operating costs.

Anthropic Leads in Business Data as OpenAI Holds Consumer Reach

Enterprise data currently gives Anthropic a narrow advantage. Ramp’s July update placed Anthropic at 42.4% among businesses in its dataset during June. OpenAI stood at 39.5%, giving Anthropic a 2.9-point lead.

Consumer forecasts show a different picture. Reuters cited an eMarketer estimate that 36.6% of US internet users would use ChatGPT in 2026. Claude’s projected share was 5.4%, indicating a much larger potential audience for ChatGPT.

Anthropic has gained ground in business subscriptions and coding tools. OpenAI retains broader public recognition and is using that distribution to move into professional workflows.

IPO Race Raises the Stakes

The rivalry is also moving toward public markets. Anthropic announced a confidential US initial public offering filing on June 1. OpenAI disclosed its own filing one week later, putting both companies in competition for investor capital.

Reuters reported that OpenAI had considered a listing as early as September and sought a valuation near $1 trillion. The company, however, did not announce a final schedule, valuation or offering terms.

The rivalry began in late 2020, when Dario Amodei left OpenAI and co-founded Anthropic. The new company pledged to prioritize safety. Reuters later reported that tensions between the companies spread to model launches, advertising and financial reporting.

Competition between OpenAI and Anthropic could accelerate product releases, while safety failures could increase regulatory pressure. Both companies have faced government intervention involving powerful models. Anthropic temporarily suspended Fable 5 and Mythos 5, while OpenAI delayed the wider GPT-5.6 rollout during a federal review.

Lawmakers have introduced separate proposals covering security audits and emergency shutdown powers. The kill-switch bill would apply to models posing serious risks during loss-of-control events. Another proposal would require independent security audits for powerful AI systems.

The open-model debate has therefore become another point in a broader rivalry over products, adoption and financing. OpenAI signed the coalition letter opposing premature restrictions on open-weight AI. Anthropic remains absent from Microsoft’s current list, though that absence does not confirm opposition. The next signals to watch are changes to the signatory list, details of any federal restrictions and how the companies apply their safety policies as more capable systems move into commercial use.