Topic
Open-weight models
AI models whose weights are released publicly for anyone to download, run and fine-tune.
Nvidia just bought the referee
$12.93bn for a company doing ~$150m in revenue. Nobody pays 86x sales for a website. They pay it for a chokepoint.
DeepSeek's benchmark chart is doing work its own harness can't back up
0813 is a genuinely good, genuinely cheap coding model that ranks last of seven on agentic coding. DeepSeek would rather you noticed the first half.
Kimi K3 is 'open' the way a container ship is portable
2.8 trillion parameters, a custom licence, and a 64-accelerator minimum. The weights are public. The capability is not.
LongCat-2.0 'beats GPT-5.5' by 0.9 points. That's not the story.
A food-delivery company trained a 1.6-trillion-parameter model with zero Nvidia chips and gave it away. Everyone's reading the wrong number.
Tokenmaxxing is over: Chinese open models are now 61% of top API traffic
The trillion-dollar labs sold you 'just spend, prices always fall.' In June it broke in public.
Open weights caught the frontier. The moat is leaking.
When you can download a near-top-tier model for free, 'exclusive access' stops being a business.
Qwen quietly won open-source AI — then half-closed the door
The most-downloaded open model family also just shipped its first closed flagship. Read the signal.
Kimi K2.7-Code: an open model that actually shipped its weights
After months of 'open-weight, weights pending', a refreshingly literal release — and a benchmark caveat you should read before tweeting the number.
Mistral wants €20bn. The valuation is the pitch.
Double the price in months — on revenue that's real but a long way from the number.
MiniMax shipped an 'open-weight' model with no weights. Bold.
Frontier claims, vendor benchmarks, and a Hugging Face link that wasn't there yet.
DeepSeek V4: the benchmark is the distraction. The chip pivot isn't.
1.6 trillion parameters, open weights, cheaper than the competition — and the real news is none of that.
Mistral's free TTS: the scary part isn't the benchmark
Open weights versus a rented-voice empire. The benchmark is the distraction; the business model is the threat.