Search: “models”

I want to prevent a race into unmonitorability kicked off by confused reporting. The depth of the computation graph for our present frontier models, including Astra, is within a factor of two of GPT-4

meta, the frontier ai lab? 1/ today we’re releasing muse spark 1.3—available in muse code & the meta model api. this is our most capable model yet—frontier performance almost too cheap to meter. much

Our partners @LangChain generate billions of tokens of agent traces a day, their richest signal on how real users react. Judging every one with a frontier closed model like GPT-5.5 or Opus is too cost

GLM‑5.3 demonstrates a notably lower hallucination rate compared with other contemporary large language models. According to a report by Together AI at TogetherCompute (published August 30 2026), Clau

Want to test how good GLM-5.3-Flash is? I have been finding it useful for visual explainers using /eli5. Try it with Pi or Hermes Agent in our new agent playground: https://academy.dair.ai/dashboard/p

Replit has introduced a free mode powered by OpenAI’s GPT‑5.6 Luna model, announced on 19 August 2026. The service offers real‑time AI assistance for coding and collaboration at no cost, aiming to bro

3000000000 downloads! Can you count the zeros at a glance? 😎 Thank you all for the incredible love. Let's keep growing together! 🌱 Alibaba's open-weight models have accumulated more than 3 billion glo

DeepSeek V4 Pro (Aug 2026) has been released on the Together AI platform. The model employs a 1.6‑terabyte Mixture‑of‑Experts architecture with a 1‑million‑token context window and supports three dist

🚀Qwen3.8-27B flies on a laptop, becoming part of our work and daily lives. Thanks for the shoutout! @atomic_chat_hq Run Qwen3.8 27B locally via Atomic Chat💥 We released Atomic Dynamic GGUF quants, fro

The benchmark for non-verifiable domains is often the opinions of humans. That is how we determine whether writing or an idea or a pitch is good in the real world And we know how to measure & benchmar

Ollama now supports the DeepSeek Harness. ollama launch dsh Run it completely in your own environment. It comes with Ollama's web search pre-installed. You can use its trajectory view to see what is h

// The Bitter Lesson of Tool Calling // Tool calling is a design choice, and the defaults are quietly costing accuracy. How so? New research releases a generation-spanning comparison of programmatic t

OpenAI announced the release of GPT‑5.6‑Cyber as part of its Cybersecurity Initiative, “Daybreak.” The model is aimed at advanced, authorized security research and testing, helping trusted defenders d

critical history and context that a lot of people have conveniently forgotten For a very long time most high-performing AI models were end-to-end neural models; vector input -> vector output, with onl

Outlook where frontier AI is headed next 18 months: The AI reasoning training + harness loop works if you can produce enough data and reasoning traces (via verifiers). Proven with code and math result

Remember OpenAI tripled its score on ARC-AGI with a harness fix? We fixed harness for Kimi K3 on CyberGym E2E: 2x better at vulnerability detection, 3x at patching K3 is SOTA cyber-defense model you c

New research from Microsoft. This one is on training computer-use agents at scale. Recent pipelines generate synthetic environments in bulk, which moved the bottleneck from how many exist to what is i

sure we lose money on every inference but we make it up in volume We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices f

This is absolutely wild... Anthropic reviewed their logs and found out that their own supposedly-sandboxed cyber evals had hacked three separate companies back in April without them noticing! In a rev

If I only went off X posts, I'd think Ramp was an AI lab We’re open-sourcing PorTAL, our framework for shared task representations and cross model LoRA adaptation. It now spans from hybrid attention m

Happy to have @FireworksAI_HQ as our day0 launch partner and bring Kimi K3 to more developers. With Fireworks, you can now deploy and fine-tune the 2.8T Kimi K3 model with just a few clicks! Kimi K3 i

Ha! It did it: 'We introduce BenchBenchBenchBenchBench (BBBBB), an executable benchmark of AI-authored conformance suites for benchmark-evaluation metrics' I really thought it would treat 'now do benc

excellent For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen

We benchmarked Gemini 3.6 Flash and Gemini 3.5 Flash Lite on document understanding. We compared against their prior versions - Gemini 3.5 Flash and Gemini 3.1 Flash Lite. 1️⃣ Gemini 3.6 Flash has rou

Kimi K3 is second only to Fable 5 on AA-Briefcase, our agentic knowledge work benchmark, but costs more than Opus 4.8 to run while averaging nearly an hour per task Last week @Kimi_Moonshot released K

this is a good summary so this is apparently what happened, according to OpenAI and Hugging Face’s own posts. wild. tl;dr: • OpenAI cyber eval – GPT-5.6 Sol and a more capable pre-release model ran Ex

75.4% SWE Bench Verified / 53.9% SWE Bench Pro on 1 bit quantisation is 🤪 This is in line with my expectations & you can expect even lower drop off with NVP4 base trained models - why not run everythi

this is a great approach, seeing this more @flymy_ai also does this when you build an agent via their api, they'll build a deterministic reusable workflow, except for where you need models we built th

One of the only times I remind people I have a PhD in computational neuroscience is when people without a neuroscience background say their model works 'like the brain.' In these cases, I put on my ne

so much for recursive self improvement, to the degree that it requires scientific taste the other thing im noticing while working on my research projects is how limited these models are GPT-5.5-xhigh

Loved the chat between @trq212 @_catwu @simonw at AI Eng summit. My top 13 takeaways from their session -> 1. Engineers should become better at product/business sense. 2. Don't worry about major rewri

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose

Human intelligence is fundamentally a collective intelligence. We solve complex problems by participating in a vast cultural network that builds upon ideas across generations. I believe the strongest

Try it out! We are seeing amazing results with GLM 5.2! it is indeed quite good! don't try it in claude code/codex - those harnesses are overly tuned for their proprietary models dcode (deepagents cod

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development 'Any topic related to building pretraining pipelines, distributed training infrastruct

arXiv:2601.14063v2 Announce Type: replace-cross Abstract: Cross-cultural competence in large language models (LLMs) requires understanding and adapting Culture-Specific Items (CSIs) across varying cul

Nex-N2-Pro is a model available on Hugging Face that can be run locally, enabling users to execute the model on their own infrastructure rather than relying on cloud services. The model is distributed

Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25+ notable open-weight drops across every modality: 🧠 LLMs → NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE,

This is a setup guide for running Nemotron-3 Ultra, NVIDIA's open-source language model, through Nous Research's platform. The guide directs users to sign up for the Nous Portal and access documentati

Nemotron 3 Ultra is NVIDIA's latest language model featuring significant improvements in speed (5X faster) and cost efficiency (30% cheaper) compared to previous versions, positioning it as a frontier

MTP means Multi Token Prediction. It's a speculative decoding technique that can result in large inference speedups in many cases. 1. Update to LM Studio 0.4.14 2. Download a model that supports MTP l

I like what Langchain recently released in their deepagents harness, which is an adapter to modify the syntax of primitive file system commands depending on the model Claude likes “Bash”, Gemini likes

DeepSeek maintains the highest efficiency in terms of active parameters relative to total model size, outperforming competitors in the ratio of parameters actually used during inference versus total t

ERNIE 5.1 is here 🚀 ERNIE 5.1 significantly reduces pretraining cost while compressing total parameters to ~1/3 and activated parameters to ~1/2 — using only ~6% of the pretraining cost compared to mo

Hello again, everyone! Our latest Qwopus3.6-35B-A3B-v1 is now live, and it is once again breathtaking! Full HF space benchmark showcase and write-up is in the comments, so you can make conclusions for

Allen AI released the OlmPool research series on Hugging Face, featuring early 7-8B parameter language model checkpoints trained on 150 billion tokens. The research explores how minor architectural mo