Altman’s email shows why OpenAI weighed a weaker local model

Simon Willison published a quote from a Sam Altman email dated October 1, 2022, exposed in Musk v. Altman. In it, Altman says OpenAI had been discussing open source strategy and wanted to create a model with approximately GPT-3 capability that could run locally on consumer hardware.

The 2022 email treats local GPT-3 as a strategic brake

The quote is short, but dense. Altman says the topic would go to the next board meeting and that OpenAI wanted to move before Stability or someone else did.

The key line is the motive: Altman wrote that such a release could discourage others from releasing similarly capable models and make new efforts harder to fund. That is not a romantic story about open source. It is market strategy in one internal sentence.

This is not a new OpenAI product announcement. It is a quote surfaced by Simon Willison from an email said to have been exposed in the 2026 Musk v. Altman case, so anything beyond the quoted text needs caution.

Open source appears here as a market signal, not a community gift

For developers, the useful reading is that local inference was not only a later answer to privacy, cost and sovereignty demands. In 2022 it already appears as a way to shape expectations: if a dominant player releases a sufficiently capable local model, investors have a harder story to tell about funding another similar team.

That also changes how current open model debates should be read. When a major AI company talks about openness, the license is only one part of the story. Timing, capability level and the startups starved by the move matter just as much.

A local GPT-3 class model then would not equal today’s open weights

GPT-3 is a capability reference in the quote, not a full technical spec. The email does not say how large the model would be, what license it would use, what data trained it or whether the weights would be fully open.

That is where restraint matters. One quote cannot prove OpenAI had a finished model or that a single person killed the plan. What it does show is that the option belonged inside a strategic discussion.

The next signal is whether archived strategy becomes current policy

The useful thing to watch is not another courtroom excerpt, but company behavior. If OpenAI or another large player releases a local model just below the frontier, the license, hardware requirements and timing against competitors will matter.

The second signal is investor reaction. If a big player’s open model chills funding for smaller labs, Altman’s email will stop being an archive curiosity. It will read like an operating manual.

Lilith's verdict

Altman’s email is a small receipt from 2022: open source can look like a gift, but sometimes it is a check slid across the table so a rival packs up and leaves.