Does Bonsai 2 27B really keep 98% of Qwen3.8 in 5.95 GB?
PrismML released Ternary Bonsai 2 27B on 17 September 2026, a 1.72-bit version of Qwen3.8 27B that fits in 5.95 GB. We read the model card, the whitepaper numbers and ...
Tag: news
Jev 1.13 bills $42 a billion tokens, and it can’t count
On 15 September 2026 TypeSafe AI put Jev 1.13 into early access: a text-in, probability-out model at $42 per billion input tokens with free output, 70 to 500 ms a ...
Qwen3.8-Omni-Flash bills audio at $0.15 and ships no weights
On 17 September 2026 the Qwen team put Qwen3.8-Omni-Flash on Model Studio at $0.15 in and $0.47 out for every modality. No weights, text output only, and a harness repo ...
AWS can’t restore me-south-1, six months after the drone strikes
On 15 September 2026 AWS said it can't restore data hosted exclusively in me-south-1 or in mec1-az2. Multi-AZ did what it was designed for, and that wasn't enough.
Gemini 3.8 Live Extended Thinking rejects any tool that blocks
Google shipped two Live models on 15 September 2026 at the old audio price. The Extended Thinking one reasons while it talks, and it errors on any synchronous function. Here ...
Atria Dawn Preview is 744B under MIT, and the BF16 weighs 1.5 TB
No launch post and no price list, but 1.5 TB of MIT weights on the 744B GLM-5.2 base, a 16-row benchmark table and a text-only API. We read all of ...
OpenAI’s Agents API has no fee, no ZDR and a one hour sandbox clock
OpenAI's Agents API puts the Codex harness behind one API call, with no fee of its own. The bill is tokens, tools and sandbox minutes, and the fine print is ...
Fugu Max costs $2 and $6 while Fugu Ultra v2 runs without Fable 5
Sakana AI released two orchestrators on 11 September 2026. Fugu Max routes over open and specialist models at $2 and $6 per million tokens. Fugu Ultra v2 keeps the $5 ...
DeepSeek V4.1-Flash arrived, and the V4 Pro retirement lasted a day
DeepSeek shipped V4.1-Flash on 10 September with a new encoder-decoder design, native vision and MIT weights, and said V4 Pro traffic would be routed into it from 14 September. The ...
SWE-2 trails Fable 5.1 by one point, and by 28 on Terminal-Bench 4
Cognition shipped SWE-2 on 10 September 2026, a Kimi K3 post-trained with RL. It sits one point under Fable 5.1 on Cognition's own benchmark and at less than half of ...