AI Model Updates: Releases, Capabilities, and Open Weights

Follow consequential developments in foundation models, multimodal capability, reasoning, and open weights.

Key updates from the last 30 days

Ordered by publication time and updated as new items arrive.

Tune the Harness, Before Tuning the Model with LangChain | Nemotron Labs

Tune the Harness, Before Tuning the Model with LangChain | Nemotron Labs nitter.net/i/broadcasts/1MJgNNeWb… NVIDIA AI Tune the Harness, Before Tuning the Model with LangChain | Nemotron Labs https://nitter.net/i/broadcasts/1MJgNNeWbklGL

Jul 21, 2026

AI Model Updates: Releases, Capabilities, and Open Weights

Key updates from the last 30 days

Tune the Harness, Before Tuning the Model with LangChain | Nemotron Labs

Upgraded versions of our Flash and Flash-Lite models are rolling out today at

Setting a World Record for MoE Pre-Training on NVIDIA GB300 NVL72 | NVIDIA Technical Blog

How Meta’s AI Models Are Powering the First Wave of Genesis Mission Projects

it is good now!

We're offering grants of up to $50,000 in Claude usage credits to researchers accelerating

AI for Science

Multimodal Interaction & World Model

Inkling from @thinkymachines is live on OpenRouter. An open-weights MoE model with 975B

Inkling from Thinking Machines Lab on Together AI. A multimodal MoE model built for

Kimi K3 may be the DeepSeek 2.0 moment. The benchmark results are out, and they are

Congrats @thinkymachines on the new open model 🙌 Inkling was trained on NVIDIA GB300 NVL72

Can't Find Benchmark and download? See here 👇https://huggingface.co/AngelSlim/Hy3-GGUF

Bonsai 27B is a big step for local AI: 27B-class multimodal capability in a phone-class

GPT-5.6 is now generally available on Bedrock

Step 3.7 Flash: multimodal reasoning at scale

Lessons From the Leaderboard: What 5,000+ Kagglers Taught Us About Improving AI Reasoning

GPT-5.6 for health intelligence, the team has been focusing hard on improvements here

GPT-5.6 is a major step forward for health intelligence. Across the lineup, we’re

50-year old math conjecture solved with Sol Ultra. Feels like the limit of what you can do

the sun is out today

GPT-5.6 is here but also the long awayited Superapp the tl;dr The model side is impressive

GPT-5.6: Frontier intelligence that scales with your ambition

UST is bringing Claude to physical AI

Ben Bernanke appointed to

Native-speed vLLM transformers modeling backend

First Look @ Gemma 4 on

Open, convenient and predictable: Introducing Provisioned Throughput

We’ve built Cohere Transcribe Arabic, the world’s most accurate open-source model for

GPT-Realtime-2.1-mini is now available in the API, bringing reasoning and tool use to our

🚀Hy3 is here. 295B MoE

Inference, optimized: How we benchmarked

What 5,000 Kagglers Taught Us About Improving AI Reasoning | Nemotron Labs

Gemma 4 on

Since June 12, we’ve been working closely with the US government to restore access to

AI training vs. inference: what's the difference?

Maximize AI Factory Energy Efficiency Through Full-Stack Inference and Training