Accelerating Open Source AI @kohai_co Puts An AI Workforce Inside Every Business
United States
Joined April 2026
- Tweets102
- Following6
- Followers1.2K
- Likes89
Pinned Tweet
Since Feb 8, Chinese open-weight models have carried ≥30% of tokens US firms route through OpenRouter, every week — peaking at 46%. Prior-year average: 11%.
Edition one of BasedAI Research, our new numbers-first series, is live:
basedai.co/research/open-wei…
Everyone is hyping open weight this week.
@Meta @nvidia @deepseek_ai all dropped new agent models.
One runs on a laptop. One runs 1M tokens at 3B active. One is 1.6T under MIT. They're not the same thing. Routing to the wrong one costs you 10x.
We route to the right one automatically. You should too.
BasedAI retweeted
Open Source AI community keeps winning.
Thanks for your contribution to the movement @finkd
Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.
Stop building agents. Start hiring a team.
AI specialists that work alongside you, not a framework you babysit @kohai_co
A team of AI colleagues uses your tools and hands you finished work. Setup, maintenance, debugging? Handled. You focus on the business and Kōhai runs the team.
Stop building AI agents and hire a team instead
Read our latest blog 👇
kohai.co/blog/stop-building-…
AI colleagues for the work that eats your week:
Outbound follow-ups that actually go out, a support inbox answered in your tone, or even the weekly numbers, pulled and explained.
They work with you, inside your tools and check before anything big ships.
We're opening to more teams soon. Get in line at kohai.co
kohai.co/blog/colleagues-not…
Welcome Qwen3.8 Max
It's clearly open-weights season!
📢Meet Qwen3.8-Max — our most capable model to date.
Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all!🎉
Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters:
- Autonomous coding: 10+ days of self-evolving development, from empty folder to production without hand-holding, complete project trace in the GitHub：github.com/qwen-code-dev-bot…
- Real work, real results: Production-quality deliverables across hundreds of professions.
- Long-horizon mastery: System-level autonomous planning with closed-loop adaptive learning, driving 500+ turns of chip design optimization and 365 days of e-commerce strategy.
- Native multimodal intelligence: Vision isn't just input — it's a continuous feedback loop for planning, execution, and self-correction.
💰Pricing:
Input: $2.0 / M tokens
Output: $6.0 / M tokens
Implicit Caching: $0.25 / M tokens
Start building with Qwen3.8-Max! 🚀
📖 Blog: qwen.ai/blog?id=qwen3.8
✅ Qwen Studio: chat.qwen.ai/?models=qwen3.8…
⚡ API: qwencloud.com/models/qwen3.8…
The biggest AI model ever is free this week. It's also 1.4 terabytes and wants 18 GPUs.
New on the blog: what open weights actually change for a small team — your costs, not your hardware.
kohai.co/blog/never-download…
Two labs bragging about second place. For most businesses, second place at a tenth of the price is first place.
New on the blog, from @TeanaTaylor
basedai.co/blog/close-enough…
BasedAI retweeted
Our @BasedAI_co team published its own look at this shift last week, The Open-Weight Turning Point.
Our data also revealed what the launch notes show:
the ability gap between open and closed models shrank from 8% to about 3%, open-weight tokens cost 60-90% less than closed-source models, and the share of work going to open models stayed above 30% since February. 🧵👇
basedai.co/research/open-wei…
Kimi K3 on Wednesday. Qwen 3.8 on Sunday.
Two trillion-parameter open-weight releases in five days, and both were introduced the same way: measured against Claude Fable 5, the best closed model in the world.
Open weights stopped asking permission to be compared. Check our latest research on the open-weight inflection
basedai.co/research/open-wei…
Since Feb 8, Chinese open-weight models have carried ≥30% of tokens US firms route through OpenRouter, every week — peaking at 46%. Prior-year average: 11%.
Edition one of BasedAI Research, our new numbers-first series, is live:
basedai.co/research/open-wei…
We’ve been saying it for weeks. Now it’s undeniable: open-weight models are beating the frontier labs.
The new era is based
Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.
This is a 17-place jump from Kimi-k2.6 (#18 -> #1).
In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.
The full model weights will be released by July 27.
Congrats to the @Kimi_Moonshot team on this major milestone!
The @linuxfoundation is launching open standards for AI token usage and billing — the meter itself, going open.
First the models. Now the measurement.
You can't govern AI spend you can only measure on a vendor's terms.
linuxfoundation.org/press/li…
Tesla capped employee AI spend at a reported $200 a week. Uber added tiers. Accenture told staff to use less.
Rational — and a confession: the price of AI work isn't under the buyer's control. Open models flip that. The fix is cheaper work, not less of it.
Yesterday was demo day at UK AI Agent Hack EP.5, Imperial College London. A week of builders solving enterprise memory governance on open weights.
Open source AI only wins if the people building on it can ship. So we put real product and real interviews in their hands, not just a prize.
To everyone who built this week, thank you. You are exactly who we want on this team.
We’re hiring.
Join us at Imperial College London for a full-day Opening Ceremony featuring an incredible lineup of AI leaders, founders, and builders, including:
Sana Wajid — CDO at @Fetch_ai, SVP at @fetch_ai_IL
Kwadwo Benko — AI Transformation Change Lead at @Microsoft
Thomas Borrel — Inference & Engineering at @BasedAI_co
Etienne Leroy — Director at @opentensor/@bittensor
Peter Carroll — Founder of @CoralOS_ai
Laura — @OpenAI
Henry Thompson — Co-Founder & CTO of @ConductAI
@EMostaque — Former CEO of Stability AI, founder @ii_posts
@CharlieMuirhead — Founder of CogX @CogX_Festival
Harry Uglow — DEX Co-Founder, a16z Scout
Paul Müller — EWOR Partner, Adjust Co-Founder
Albert Phelps — tomoro.ai
Hazel Hu — Team Lead at @GCCofCommons
Bart Zuber — Building AI at Vega
Romain Billot — Core Contributor at Kaspa
We’ll have keynotes, panels, sponsor sessions, builder guidance, and a full agenda designed to help hackers get ready for the week ahead.
Hackers will also receive 1 month of Codex to support their builds during the hackathon.
And to close the day, we’ll host the @cantor8 Dinner & Co-working session — a nice meal, founder networking, and a £7,500 cash prize pool. The first 150 participants will each walk away with at least £50 cash prize.
Date: 28/06/2026
Venue: CAGB 200, Imperial College London
Sign up and view the full agenda here: aiagentslab.uk/hackathon/ep5
New on the blog
Routing, Not Rationing 👇
basedai.co/blog/routing-not-…
The @coinbase AI-cost memo is making the rounds. Strip the company name off it and it’s a clean blueprint for the next two years. Here’s the honest version of what it actually says.
Routing beats picking. Preprocess the task, send it to the model that fits. A frontier model for the hard planning step, an open model for execution where frontier is overkill. Most jobs are execution. That’s where the bill — and the saving — actually lives.
And caching beats everything. Their cache-hit rate went 5% to 60% once requests were cache-aware. None of this is “use the cheapest model.” It’s “build the layer that routes.” Open weights make it economic; routing makes it disciplined. That layer is what we build at BasedAI — BasedAPIs for the engineers who want to call it, @GetHirebased for the teams who just want the work done.