August was the month the competition between OpenAI and Anthropic got direct.
Anthropic Cuts Off OpenAI
At the start of the month, Anthropic revoked OpenAI's access to Claude. Anthropic said OpenAI's technical staff had been using its coding tools ahead of the GPT-5 launch, which its terms of service prohibit when the goal is building a competing product. It kept OpenAI's access for benchmarking and safety evaluations.¹
GPT-5
On August 7, OpenAI released GPT-5. It combines OpenAI's fast models and its reasoning models into one system, and it became the default for every ChatGPT user, including free users.² OpenAI says it is state of the art in several areas, including coding, the use case where Claude has built its enterprise lead. Coding is the most valuable AI use case right now, and both companies know it.
The $1 Government Deals
On August 6, OpenAI offered ChatGPT Enterprise to the entire federal executive branch for $1 per agency for a year.³ Six days later, Anthropic offered Claude to all three branches of government, executive, legislative, and judicial, for $1.⁴ Both companies already hold Pentagon contracts worth up to $200 million each, awarded in July.⁵
Nobody makes money at $1. The bet is that government adoption convinces large enterprises the tools are ready for sensitive work.
Testing Each Other's Models
On August 27, the two companies published results from a pilot in which each ran its own safety and alignment evaluations on the other's models.⁶ Anthropic found that OpenAI's o3 reasoning model behaved better than Claude Opus 4 on most of the dimensions it tested. It also found GPT-4o, GPT-4.1, and o4-mini much more willing than Claude or o3 to help with clearly harmful requests.
This is the most useful thing to come out of the rivalry so far. Competing on capability while cooperating on safety testing is a pattern I hope sticks.
The Money Behind It
OpenAI is in talks to sell employee shares at a $500 billion valuation.⁷ Menlo Ventures' mid-year survey put Anthropic at 42% of enterprise code generation, twice OpenAI's share.⁸ The fights this month were over real revenue.
What It Means for Product Teams
- Don't depend on one model vendor. Access can change for business reasons as well as technical ones.
- Keep switching costs low. Put the model behind your own interface and keep your evaluations portable.
- Read the cross-lab safety results. They give buyers a better view than either lab's claims about itself.
Are you building on one model provider or several? I'd like to hear how teams are handling vendor risk.
Sources
¹ TechCrunch: Anthropic cuts off OpenAI's access to its Claude models
² TechCrunch: OpenAI's GPT-5 is here
³ CNBC: OpenAI is giving ChatGPT to the government for $1
⁴ GSA: OneGov deal with Anthropic
⁵ CNBC: Anthropic, Google, OpenAI and xAI granted up to $200 million from DoD
⁶ Anthropic: Findings from a pilot Anthropic-OpenAI alignment evaluation
⁷ CNBC: OpenAI in talks about share sale at $500 billion valuation