Mozilla: The state of open source AI

Mozilla says open AI is winning, but readers are raging about the site itself

TLDR: Mozilla says open AI is getting cheap, powerful, and important for communities that want control over their own tools. Commenters, however, were far more fired up about the hard-to-read website and messy design, with a side of dramatic predictions that open tools could crush today’s AI giants.

Mozilla came in swinging with a big, idealistic message: open AI should belong to everyone, not just giant companies with giant bills. Its report argues that cheaper, more open tools are catching up fast, with some now nearly as capable as the best locked-down systems for everyday tasks. The pitch is emotional too: from Māori language tools in New Zealand to offline crop disease help for African farmers, Mozilla says open AI is about communities owning their own tech instead of renting it forever.

But in the comments? Absolute chaos. Readers barely made it to the argument before turning the whole thing into a roast of the website itself. One of the loudest complaints was that the page is a visual nightmare: giant text, weird animations, scrolling gimmicks, and a font that one person said felt downright "aggressive." Another fumed that Mozilla, of all groups, should know better on accessibility, basically accusing it of chasing trendy design while forgetting the people actually trying to read the thing.

And then came the sharper knives: some commenters called the whole write-up "slop," saying the opening was confusing and stuffed with vague slogan-like lines instead of clear explanation. Still, not everyone was just there to boo the typography. One spicy hot take predicted open models could eventually wipe out OpenAI and Anthropic, because if big cloud companies and Apple can run similar tools themselves, why keep paying the middleman? So yes, Mozilla wanted a rallying cry for open AI — but the crowd mostly showed up to drag the font, clown the UX, and argue over who gets vaporized next.

Key Points

  • Mozilla’s July 2026 report argues that open-source AI has become viable for many real-world workloads and is no longer merely a compromise versus closed models.
  • The article says the open-versus-closed capability gap on Chatbot Arena narrowed from 8.04% to 0.5%, then widened to 3.3% by March 2026 as closed reasoning models pulled ahead.
  • Mozilla reports that GPT-4-class inference cost fell from $20 to $0.40 per 1 million tokens over 36 months.
  • According to the article, open models are near parity in coding, instruction-following, and general knowledge, while closed models still lead in reasoning, long-context retrieval, and agentic tasks.
  • The report states that open-weight models now account for a majority of production tokens routed through OpenRouter, and that the platform’s five highest-volume models are all open.

Hottest takes

"This new trend of content appearing while scrolling down is so terrible accessibility-wise" — hypfer
"Maybe its the wildfire smoke in my eyes, but that font choice feels aggressive" — Cuuugi
"open models is what will kill Anthropic and OpenAI" — babblingfish
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.