June 26, 2026
Open vs closed: nerd war erupts
The gap between open weights LLMs and closed source LLMs
AI fans are fighting over whether the free models are really catching the secret ones
TLDR: One chart suggests publicly available AI could match private AI by late 2026, but a broader look says the gap may stay around five months. In the comments, readers argued over misleading labels, questionable graph logic, and whether “free” AI survives only because big companies allow it.
A spicy prediction lit up the comment section: one chart says publicly available AI could catch the big private models by December 3, 2026. Naturally, the article itself jokes that this is the moment to cash out your pension and hide on an island. But the real action is in the replies, where readers immediately slammed the brakes on the hype train. The loudest complaint? Words matter. One of the top comments flatly says the piece mixes up “open source” with “open weights” — basically, saying these models may be available to use, but that doesn’t mean the full recipe is truly public. For a certain kind of internet reader, that’s not a small typo; that’s a full-on courtroom drama.
Then came the math nerd pile-on. One commenter compared the whole “the gap is shrinking!” argument to Achilles and the tortoise, warning that trend lines can create fake inevitability. Translation for normal people: just because one line is moving toward another doesn’t mean a dramatic catch-up moment is guaranteed. Others weren’t buying the rosy future for a different reason: today’s “open” AI boom depends heavily on rich companies choosing to share their work. As one commenter put it, the spigot can be turned off at any time. Another wondered if free models are secretly living off scraps from the private giants they claim to be chasing. And in the middle of all this philosophical combat, one exhausted soul delivered the most relatable review of all: “at first glance, these graphs are confusing.” Honestly? Mood.
Key Points
- •The article measures the open-weight versus closed-source LLM gap by calculating how many months earlier the closed-source frontier reached the same benchmark performance level.
- •On the Artificial Analysis Intelligence Index, the gap began shrinking around summer 2024 and a fitted trend line projects it would reach zero on December 3, 2026.
- •The author says a single benchmark does not provide a complete picture of LLM capabilities.
- •The analysis was repeated across 18 Artificial Analysis benchmarks, using monthly box plots, average gaps, and a line of best fit.
- •Across the broader benchmark set, the average gap stays nearly flat at just under five months, with coding improving the most while most other datasets show moderate gap increases.