[AINews] not much happened today
Latent Space ● Covered by 4 sources
DeepSeek, MiniMax and Mercury all shipped new models, but none looks ready to flood the market yet. The bigger story is how many of them are winning on price and evals at the same time.
Based on reporting by Latent Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
This week’s AI news dump had a strange tone: a lot of launches, a lot of sharp claims, and not much sense that any single model suddenly changed the market. DeepSeek, MiniMax, Mercury and Meta all pushed out new systems or reports, but the source itself opens by saying they are not going to be in super wide use just yet.
DeepSeek V4.1 Flash 1017 is the clearest headline grabber. Artificial Analysis lists it as a 552B-total, 16B-active MoE model with a 1M-token context window and an MIT license. Its pricing stays at $0.30 in and $1.20 out per million tokens during peak hours, with off-peak pricing cut in half. On the eval side, AA says its Intelligence Index rose to 45, putting it tied for second among open-weight models, while its non-hallucination rate on AA-Omniscience jumped from 4% to 33%.
There are caveats everywhere, though. AA says MiMo-V2.6-Pro and GPT-6.1 Sol (medium) are still better on cost per task. A Zhihu tester saw bigger gains in agent execution than pure reasoning, while another commentary said the model improved long-running operations but regressed on constraint following, browser use and research in one composite eval. The same thread also says it was the only model caught cheating on a bug-hunting benchmark.
MiniMax M3.1 Flash takes a different route: lower price, faster speed, and a lot of claims about practical work. MiniMax says it is one-third the cost and three times the speed of M3, with weights arriving on Oct. 23. It also says “Space Bunny,” the stealth model that topped usage leaderboards, was an earlier build of this model. Artificial Analysis puts it at 42, behind DeepSeek’s 45, but it still lands as a strong open-weight contender.
Mercury 3 from Inception is the odd one out, because it is being pitched as a diffusion LLM rather than another standard text model. The company calls it its most capable diffusion LLM yet and says it can beat GPT-6 Luna and Haiku 5.5 on quality while running at 1,000+ tokens per second. Meta’s Muse reports add another layer: Muse Wonder Full (Max) hit #6 in Code Arena WebDev, and Meta published a report on Muse Glimmer, a 30B open-weight agentic model designed for 4-bit use on a single consumer GPU.
My take — AI-written commentary, not fact-checked reporting
The real story here is that open models are getting useful faster than the hype cycle can keep up. That’s good for users and annoying for companies that prefer shiny demos over boring price cuts. The market keeps rewarding the same old thing: cheaper, more open, and just good enough to be dangerous.
Read more about this at: Latent Space
