Models
New and updated models, and how they score.
Qwen/Qwen-Image-2.1-Turbo
Qwen published the model Qwen-Image-2.1-Turbo on Hugging Face.
Microsoft's Decision-1 model enters the fast-growing AI decision model race
With Decision-1, Microsoft enters the growing decision model space.
Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash
Following last week’s release of Clef and Clef-flash, Cloudflare’s open-weight decision models, we decided to bring forth more gifts.
ConwayResearch/Underdog-Saluki-27B-1.0
ConwayResearch published the model Underdog-Saluki-27B-1.0 on Hugging Face.
mistralai/Voxtral-Mini-4B-Realtime-Arabic
mistralai published the model Voxtral-Mini-4B-Realtime-Arabic on Hugging Face.
[AINews] Reflection Beam - 501B-A23B American Open Model
It’s been over a year since Reflection launched with us with big goals on coding (and hinted about their RL approach):
Understanding Jev, the new model everyone is talking about
On September 15, the startup TypeSafe AI came out of stealth and released a new AI model.
nvidia/PixelUMM
nvidia published the model PixelUMM on Hugging Face.
OrcaRouter Releases OrcaCyber Zero 1.5 Cybersecurity Model With 1M Context
OrcaRouter has released OrcaCyber Zero 1.5, a model for authorized vulnerability research.
VideoStop Renting Intelligence: The Train-to-Deploy Loop for Specialized AI — Fireworks AI
Jetashree Ravi, who leads part of the applied machine learning team at Fireworks AI, explains how teams move from closed models to open ones without losing quality.
llm-mistral 0.16
Adds support for reasoning models, such as the newly released Mistral Large 4.
Multimodal open d1 decision models for the edge
Best decision model under 10B on the Decision Index 0.2.1: d1-3B scores 48.57, ahead of every 4B and 9B model and of Decider 35B-A3B (47.11).
Can AI automate AI R&D yet?
Existing evidence shows that AI can perform software engineering tasks relevant to AI research, dataset creation, open-ended optimization of defined metrics (autoresearch), and more.
Scale Bitwise-Deterministic Pretraining with NVIDIA Megatron Core
Bitwise determinism makes large-scale pretraining easier to debug, validate, and resume reproducibly.
OpenPilot
Open-source desktop AI agent for any model you choose
Introducing GPT-6.1 Sol
Meet GPT-6.1 Sol: near-Astra intelligence for coding, computer use, and professional work at one-fifth of Astra’s standard API input and output token prices.
Last Week in AI #345 - 5 new models, 9 misalignment incidents, some Dots
Anthropic and OpenAI race to release smarter and cheaper models
Best practices for Amazon SageMaker HyperPod administration and governance
Amazon SageMaker HyperPod gives machine learning (ML) teams access to large pools of accelerated compute for training and fine-tuning models.
Sonnet 5.5 is worth a try
I’ll be at OpenAI DevDay today.
Qwen3.8 Flash Next Reasoning Modes: Off vs Low vs Medium vs Xhigh
Like Qwen3.8 27B, Qwen3.8 Flash Next has three reasoning efforts: low, medium, and xhigh.
microsoft/AesCode-32B
microsoft published the model AesCode-32B on Hugging Face.
ibm-granite/ensemble-forecasting-with-ibm-granite-time-series
ibm-granite published the model ensemble-forecasting-with-ibm-granite-time-series on Hugging Face.
VideoWhich is The Best Qwen3.8-27B?
Which is the best fine-tune of Qwen3.8 27B that reduces the amount of thinking and reasoning tokens but still keeps the best accuracy for your particular use case?
ReviewBench: An open benchmark for AI code review
Agentic code review is becoming an essential piece of how development happens.
CoreWeave Forge aims to speed up the AI improvement loop
The post CoreWeave Forge aims to speed up the AI improvement loop appeared first on SiliconANGLE.
VideoWhy Deep Learning Failed on Tables for a Decade - Frank Hutter
Frank Hutter, co-founder of Prior Labs, on why deep learning struggled with tabular data for a decade: tables are messy and heterogeneous, hyped models like TabNet did not generalise to new datasets, and there was no ImageNet of tables.
mistralai/LIDstral-Arabic
mistralai published the model LIDstral-Arabic on Hugging Face.
nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2
nerkyor published the model Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2 on Hugging Face.
unsloth/embeddinggemma-2-GGUF
unsloth published the model embeddinggemma-2-GGUF on Hugging Face.
alesha-pro/Qwen3.8-Flash-Next-abliterated-GSQ-RCO-Strata-GGUF
alesha-pro published the model Qwen3.8-Flash-Next-abliterated-GSQ-RCO-Strata-GGUF on Hugging Face.
LiquidAI/d1-omni-600M
LiquidAI published the model d1-omni-600M on Hugging Face.
perplexity-ai/pplx-decider-v1.1-27b
perplexity-ai published the model pplx-decider-v1.1-27b on Hugging Face.
VideoSame Model, Different Speed: Why Your Inference Provider Matters — FriendliAI
Open-weight models are good enough now.
nvidia/PixelDiT2-ImageNet
nvidia published the model PixelDiT2-ImageNet on Hugging Face.
VideoAgents That Own Their Inference — Du'an Lightfoot & Khaja Omer, Akamai Technologies
Speculative decoding drops a demo model from roughly 58 tokens per second to 16 instead of speeding it up.
VideoHill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Shubhankar Srivastava uses that uneven progress to show how browser agents can learn a task without changing model weights.
Aleph-Alpha/Kolibri-1
Aleph-Alpha published the model Kolibri-1 on Hugging Face.
AI agent teams waste massive tokens for barely measurable quality gains, research finds
Teams of AI agents barely outperform solo agents but cost up to 5.1x more, according to Vals AI.
VideoVector Isn't Enough: Hybrid Search & Retrieval — Jeff Vestal & James Williams, Elastic
Jeff Vestal and James Williams work backward from that agent behavior to the retrieval layer in Elasticsearch.
VideoFrom Zero to Leaderboard: Agent Evaluation — Wolfram Ravenwolf, Weights & Biases
An agent repairs a broken git repository and earns a perfect score, but the run stays out of the default leaderboard because it tested only one task.
[AINews] not much happened today
AINews’ website lets you search all past issues.
Mistral Large 4
llm -m gpt-6.1-sol 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars'
One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
Starting from Nemotron 3, our teams used supervised fine-tuning (SFT), reinforcement learning (RL), and feedback-driven inference to create systems that reached gold-medal level at both IMO 2026 and IOI 2026.
Holo4: powering generalist computer-use agents
Holo4 is our new series of agentic models.
Impactful scheduling for GPU clusters
On the AI Infrastructure team at Ai2, we’re responsible for providing the institute’s GPU compute capacity, specifically targeting large, distributed training workloads.
The model that didn't exist, so you made it yourself
Last week, I wanted a small version of the prompt rewriter that ships with Qwen-Image 2.1.
EmbeddingGemma 2
My comment on EmbeddingGemma 2 — Hacker News.
Qwen3.8 27B Xhigh or Flash Next Medium
In my Qwen3.8 27B reasoning benchmarks, Xhigh generally produced the strongest results, but it needed much longer responses than Medium.