Next-Gen Models Ready — Vendor Evaluations Built on Sand

Anyone evaluating AI vendors right now is making decisions based on models that won't exist in a few months.

That's not an exaggeration. Last week, a CMS misconfiguration at Anthropic made an internal document public — a draft blog announcement for a new model called "Claude Mythos." Anthropic confirms to Fortune: "a step change, the most capable we've built to date." A completely new model class above their current top model.

Meanwhile at OpenAI: CEO Sam Altman describes internally a fully trained model with codename "Spud" as "a very strong model that will really accelerate the economy." Release: in a few weeks.

Two companies. Two fully trained next-gen models. Both holding back the release.

The pattern behind this is revealing: Both are facing billion-dollar funding rounds — OpenAI's IPO preparation, Anthropic's next capital round. Next-gen models are the strongest signal to investors. The timing is no coincidence.

For C-level decision makers, this means three things:

→ Vendor evaluations based on today's benchmarks have a half-life of weeks. Anyone deciding today that "Model X is better than Model Y" is comparing products that both vendors already consider obsolete.

→ Enterprise contracts being signed now are based on models that both vendors internally already consider the previous generation. Anyone negotiating terms and conditions should know what's sitting in the pipeline.

→ The real question isn't which model is better. The question is: Are data, processes, and interfaces set up so the company can actually benefit from better models?

If your AI readiness isn't in order, even the best model in the world won't help you. And if you're basing your evaluation on today's benchmarks, you're building on sand.