Trump Admin Approves Limited Release of Mythos and GPT 5.6
Washington D.C. is destroying America's AI industry, turning it into a quasi public utility, thereby slowing innovation and driving prices higher.
If you scan TEK2day Pulse’s headlines (screenshot below), they are dominated by the U.S. Government allowing limited release of Anthropic’s Mythos model and OpenAI’s GPT 5.6 to a limited list of approximately 100 Enterprise customers.
This is an AI tax on small companies. Small companies like TEK2day heavily leverage AI tools to punch above our weight. Pound for pound, we get more ROI out of AI tools than large enterprises because we can train up faster and deploy AI in our own products more quickly than large Enterprises can.
The Federal Government is clearly favoring its large corporate donor class.
The reaction has been a shift to opensource models, most of them Chinese. I am playing with GLM 5.2 via openrouter, but may just run it locally. We are not the only company working with opensource models. This opensource model shift started to happen earlier this year.
My guess is that the Anthropic and OpenAI IPOs probably will not happen this year.
If they do take place, the IPOs will be priced at far lower valuations.
If those IPOs do not happen this year, my guess is that both Anthropic and OpenAI will be sold.
Anthropic to Google? I can imagine Google swallowing Anthropic given Anthropic’s ties to DeepMind (Dario Amodei worked with DeepMind CEO Demis Hassabis previously at Google and Google needs to fix the fact is does not have a quality coding model as Gemini is awful).
I am not so sure about OpenAI’s future. I can see Sam Altman being pushed out, Greg Brockman taking over as CEO, and NVIDIA, Softbank and Oracle taking big equity stakes, particularly NVIDIA.
What about the hyerscalers?
Amazon will be a cloud player without a coding agent (Kiro is crap). Amazon talked its way into becoming an LLM gatekeeper when CEO Andy Jassy complained to the Trump Administration that Anthropic’s Mythos model could be jailbroken.
Microsoft will not have a coding agent (Copilot is crap) but will also be an LLM gatekeeper.
Meta will not have a coding model, despite Zuckerberg logging employee keystrokes in an effort to build a better model. In fact, Zuckerberg and Musk (Grok, SpaceX), have spent all of their goodwill in the Bay Area and can’t hire the people they need to build a great coding model.
Automating the coding function is where it is at. This has not changed. Coding is where the money is. Coding is not “solved” despite the fact that Anthropic says that coding is “solved”. Not even close. AWS and Facebook are shipping buggy software to users precisely because they don’t review the slop the coding agents produce as one should review agentic coding output.
Agentic coding will get better. Mythos for example (as well as Codex 5.6), runs on NVIDIA Blackwell chips, not the current Vera Rubin chips.
