We built a new way to train contextual embedding models, which encode each chunk of a document with the whole document in view. pplx-embed-v2-context-9b-preview sets a new state of the art on ConTEB and @turbopuffer's new, privately held context-bench. https://www.perplexity.ai/hub/blog/contextual-embedding-beyond-the-gold-passage

Introducing Boreal-H3 — a video model built for ads and our next step toward recursive self-improvement in video generation. A good-looking video isn’t enough. The product has to stay the same. The actor has to stay the same. The label has to be right. And the action in the brief actually has to happen. So we post-trained MiniMax H3 specifically for advertising. But this isn’t a one-off SFT or LoRA fine-tune. We built a closed-loop system that learns what to improve next. Human-calibrated evaluation diagnoses failures and guides the next intervention: targeted data collection, reinforcement learning, or inference optimization. When the feedback is unreliable, we revise the evaluator or reward—not just the generator. Every experiment feeds into shared memory, informing the next training decision. The model improves, and so does the process that produces its successor. The results: → 85.3% reference fidelity — highest among the frontier video generation models we evaluated → Brief success: 28% → 50% → Identity match: 83% → 94% → Visible defects per clip: down 70% → Generation time and estimated cost: down 20% Boreal-H3 doesn’t just make better-looking video. It makes more usable ads. Credit to the @MiniMax_AI team for the foundation we’re building on. This launch is a checkpoint, not the finish line. We’re building more than a better video model. We’re building a system that learns how to make the next one better.

Introducing Cohere Embed 5: our new state-of-the-art family of embeddings models. Get frontier capabilities with Embed 5 Pro or low-latency performance with Embed 5 Fast.

Qwen3.8-27B is now live on Nebius Token Factory. A compact 27B dense model for coding, research, and agent workflows, with a focus on planning and completing tasks across multiple steps. Start building: https://tokenfactory.nebius.com/endpoints?modals=endpoint-details&model-id=Qwen/Qwen3.8-27B

New historic NanoGPT record at 39.9s (-27.7s) from @DevenPzak , obliterating the prior record of 67.6s! This record introduces a new paradigm of thinking to NanoGPT: instead of optimizing matmuls or adding more expressive operations, optimize at the individual flop level with incredibly clever engineering and ML judgement. If a flop is low value on a particular step, skip it. Specifically: -(~8s) Sampled softmax. If a token doesn’t appear in a batch, skip its lm_head fwd/bwd some fraction of the time. -Sparse values. Only run an optimizer step for ngram embeddings that occurred in the batch. Set beta1 to zero to enable this. Beta2 is applied retroactively when the row is later used. -Sparse updates. Only update ngram and value embeddings once every 4 steps instead of once every 2. -Sparse communication. Shard the n-gram table across GPUs, and only pass the rows receiving updates on each step. -Sparse optimizer states. For the n-gram table, reduce from 2 floats in Adam optimizer per param, to 1 float per 768 params. -Hand-rolled flash attention for 64 dim heads. There are several additions that add accuracy too: -(~4s) EMA during last 300 steps, combined with lifting final_lr to 0.3 instead of 0.15. -(~1s) A new optimizer, Anvil2, which expands muon via a second tracked momentum buffer, improves the ortho coefficients, and modifies the cautious weight decay application. -A couple additional dynamic skip connections in the network. The most striking consequence of the ‘flop aware paradigm’ is you can grow parameters arbitrarily large, only limited by the available memory, since you can selectively choose how to expend flops on those parameters on each step. NanoGPT has kept active parameters below 124M, but total is unbounded, and has grown to 640M through embedding sparsity over the last year. This PR takes that to its logical conclusion on the 8xH100, scaling up to 65B sparse embedding parameters, which accounts for 25% of the PR’s gains. At frontier scale, where one is not bounded by an 8xH100, one could imagine where this paradigm could lead. https://github.com/KellerJordan/modded-nanogpt/pull/360 As this was a very notable PR, I spoke with Deven for an hour to learn how he did it. Here’s his story on the changes: https://hyperstition.cc/training-nanogpt-in-39-9-seconds

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

MiniMax's latest text model, M3.1-Flash-Preview, debuts today on MiniMax Code. Built for everyday development, it's fast, reliable, and ready for real work, from quick bug fixes to full features.

was looking for a quiet weekend but xiaomi dropped their rl envs repo last night to put in perspective, if you have to buy some tasks like this its usually hundred to thousand dollars per task so this repo is literally worth millions https://huggingface.co/datasets/XiaomiMiMo/MiMo-V2.6-RL-oss