CTO @mozilla. board: @tumocenter & @medic. fmr CTO @EmCollective & director @uber ATC (🤖 🚗)

Bay Area
this is just a slice of a much longer conversation. the full version goes further: piped.video/watch?v=eaUCbH0m…
Open source models are rapidly narrowing the gap between closed ones and the concerns are real. At the same time, people are building ethical, human-first efforts. It all comes down to how AI is built, audited and used. This is just one of the conversations from the latest Owners Not Renters livestream featuring Alejandro Gonzalez and Thomas Bille, and Brian Mbuba. Watch the replay and stay tuned for more information about our next live show.
384
open models are free to download and you can take them anywhere. so the meter moved to the things you can’t take with you when you leave: the cache, your memory, and the gateway. when one company owns the router and the payment rail, it knows exactly how much it would cost you to leave. run a router you control. route on cost. keep a second model warm. stateofopensource.ai
2
4
965
Raffi Krikorian retweeted
Per-token price is the wrong way to compare AI models for agentic work. Kimi K3 lists at $15 per million output tokens. On the Artificial Analysis Index it spends 130M output tokens against a 63M median, so on the same suite at the same list rate it bills like $31/M. The problem with ranking by per-token price is that it hides a 2x spread in what a task actually costs.
4
4
25
2,752
want to build your own AI stack? tomorrow you'll get to to hear from the people who ship the tools. @MozillaAI's Alejandro Gonzalez and Thomas Bille, plus Brian Mbuba, join the next Owners Not Renters livestream to show how. join host @Marcus_Rein_ tuesday, oct 13 at 11am ET. bring your questions 👇 streamyard.com/watch/Mz78rUH…
2
317
Raffi Krikorian retweeted
TODAY Fri. 9 Oct 2026, LIVE, 1 pm ET / 10 am PT Open Source & Open Weight AI, Rent or Own? Join #CXOTalk guest @Raffi, CTO, @Mozilla. cxotalk.com/episode/mozilla-… What’s holding back US development of competitive #OpenSource and #OpenWeight models? #CTO #CIO #AIagents #FutureofAI
2
3
191
capability gains in AI no longer require a new model. the model and the harness still matter, but real results happen when working on an open base. in july, DeepSeek ran a post-training pass that lifted V4-Flash 21 points on Terminal Bench and 10 on the  AA index. same model. same architecture. same size. stateofopensource.ai/
2
2
2
867
check the economics. post-training costs a fraction of pre-training. take @thomsonreuters, which recently trained a Qwen-based model for $450,000. that’s a cheaper path to capability gains than switching out models. run this on an open model and you hold the weights, the adapters, and the reward functions. all this and more in the full @mozilla State of Open Source AI report stateofopensource.ai/
141
an agent can prove who it is. no one can say what it’s allowed to do. MCP went from ~2M to 110M+ monthly SDK downloads in 21 months. meanwhile ~21% of companies say their agent governance is mature. stateofopensource.ai/
3
2
5
1,023
MCP now requires OAuth 2.1 + PKCE. A2A ships signed Agent Cards. both answer "is this agent who it says it is?" neither answers what a verified agent may write, or where the data it touches can go.
1
179
this is already at production scale. 28% of the Fortune 500 were running MCP in production by march 2026. what's missing is portable permission: rules that travel with an agent across tools. if an open standard doesn't set them, closed platforms will. full data in Mozilla’s state of Open Source AI report stateofopensource.ai/
129
join me live this friday at 1pm ET/10 am PST. link to join below 👇 #CXOTalk
Who owns your AI stack– you, or the vendor that you rent it from? Mozilla CTO @raffi Krikorian will be a guest on @cxotalk with @mkrigsman on October 9 at 1pm EST / 10 am PST. He’ll discuss the difference between open source and open weights as well as some of the risks and limitations associated with open source AI. Join live: cxotalk.com/episode/mozilla-… #CXOTalk
385
serious AI used to live on someone else’s data center. you rented it by the token. that’s changing. Capable open models now run at every hardware tier, from a gaming PC to a small data center. top models improved by adding hardware. smaller models improved on the same hardware. capability exists at every tier. the complete overview is in the State of Open Source AI report: stateofopensource.ai/
2
2
13
2,006
the top tier bought its gains with more hardware. the lower tiers gained points on hardware that stayed fixed.
3
140
every closed model can be changed, repriced, or switched off by someone who isn’t you. this week i spoke at @CTOCraft Europe in Amsterdam. i asked them what their plan was for the day their AI provider changes right under them. most don’t have a plan. they have an untested back up plan. and it’s where most AI stack currently sit. here's the big takeaway: you don’t have to own everything. you have to be able to leave.
1
2
355
Raffi Krikorian retweeted
Once open models get within a few points, benchmarks may not matter as much as optionality. Open weights give control back to the user. Rent a closed model, you’re locked in and the price is theirs to change. Self-host on infrastructure you own, and no one can repossess it. Where do you want to build? Read the full State of Open Source AI report → stateofopensource.ai
11
9
38
3,995