ALL INPredictions
Last updated Nov 29, 2025
Prediction
Chamath
← Back to hostEpisode
Chamath @ 00:24:23Right
aitech
Within a few months of September 2024 (by early 2025), OpenAI will release a production version of its O1 reasoning model, which will be a materially more capable, "spectacular" upgrade over the preview version.
OpenAI will preview O1 and then they'll have the actual O1 production build probably in the next couple of months, which will be probably pretty spectacular.View on YouTube
Explanation

Evidence from OpenAI’s own materials shows that this prediction basically came true on both timing and substance.

  • Preview then production: OpenAI announced OpenAI o1‑preview (with o1‑mini) on September 12, 2024 as an early preview reasoning model in ChatGPT and the API. (openai.com) Later, OpenAI documentation for ChatGPT states that “ChatGPT Enterprise and Edu customers will have access to the o1 model on December 5”, referring to the non‑preview o1 model. (help.openai.com) That’s about 2½ months after the September 20, 2024 podcast—within a “couple of months” window.

  • Production build: In the product blog “OpenAI o1 and new tools for developers,” OpenAI describes o1 (not o1‑preview) as “the successor to OpenAI o1‑preview” and explicitly calls it “production‑ready”, listing key production features like function calling, Structured Outputs, developer messages, vision, and a reasoning_effort control. (openai.com) The same article notes that the snapshot o1‑2024‑12‑17 is the version being shipped in the API, clearly marking it as the production release of the o1 line. (openai.com) The API pricing page separately lists o1 as a regular, billable model, distinct from preview models, further confirming production status. (platform.openai.com)

  • “Spectacular” / materially more capable: The o1‑2024‑12‑17 snapshot sets state‑of‑the‑art results on several benchmarks and is significantly stronger than o1‑preview—for example, AIME 2024 accuracy jumps from 42.0 to 79.2, MATH from 85.5 to 96.4, and SWE‑bench Verified from 41.3 to 48.9—while also using ~60% fewer reasoning tokens and adding major capabilities (vision, function calling, structured outputs). (openai.com) These are large, tangible gains over the preview model, consistent with a “spectacular” or materially more capable upgrade.

Given that: (1) a non‑preview, production‑ready o1 shipped around mid‑December 2024, roughly “in the next couple of months” after the podcast, and (2) it is clearly a significantly more capable successor to o1‑preview, Chamath’s prediction is best classified as right.