New models and versions, open weights, and changes in capability and pricing, with their evaluations.
StepFun's flagship base model Step 5 Preview reached the top of OpenRouter's Trending list and second place on its daily ranking, with an official open-source release set for October 15. The sparse MoE model has 600 billion total parameters and about 27 billion activated per token, and Artificial Analysis gives it a composite intelligence index of 44, placing it in the global open-source top three.
Falcon-Emirati-7B is a specialized model for Emirati dialects built on Falcon-H1-Arabic. It aims to understand and generate Emirati Arabic like a native speaker. It uses the Falcon-H1 hybrid architecture, running Mamba alongside Transformer attention. The base model family comes in 3B, 7B, and 34B sizes, with a context window of up to 256K tokens.
Reflection released Beam, a text-only mixture-of-experts model for coding, agents, and scientific work. It has 501B total parameters and 23B active parameters, was trained from scratch, and will release all weights under Apache 2.0 later this month.
Reflection AI officially released its first frontier open-weight model, Beam, saying it performs comparably to leading Chinese open models on advanced reasoning benchmarks with 3–4 times less inference compute. Beam is a 501-billion-parameter, 23-billion-active-parameter text model, pretrained on 23.8 trillion tokens with a 1-million-token context window. The company says it matches Z.ai's GLM-5.2 and exceeds current leading Western open models.
Aleph Alpha released Kolibri, a bilingual German-English model with 78B total parameters using a mixture-of-experts architecture that activates about 3B parameters per token. The company says Kolibri sits on the Pareto frontier for quality and operating cost in both languages. It scores 71% on German benchmarks and decodes faster than comparable models including GPT OSS A5B, Qwen 3.6 A3B, and Gemma 4 A4B.
NASA and IBM Research, together with several academic institutions, released the NASA-IBM Lunar Foundation Model, which they describe as one of the first open-source foundation models for lunar science.
Ai2 open-sourced AstaBrief 8B, a model that turns research questions and retrieved literature excerpts into reports with citations. It was trained from Qwen3-8B using SFT and DPO and is now available as Asta's Fast mode.
Why it matters: The source describes AstaBrief's training-data construction and citation-filtering details, which can be adapted to deploy other long-form generation models.
AI2 released Olmo-core 3, a major upgrade to its large language model development framework. It redesigns the open MoE training system to scale MoE training to trillions of parameters while maintaining compute efficiency.
Why it matters: The source provides throughput and memory comparisons for the MoE training stack, helping readers assess changes in the cost of large-scale MoE training.
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
Google DeepMind launched SynthID Bio, bringing watermarking to synthetic biology. It embeds imperceptible signatures directly into biological code so that generated content can be verified both in digital models and in synthesized physical proteins, while preserving their biological function.
Why it matters: SynthID watermarking has expanded from images to synthetic biology, showing how AI-generated proteins can be marked and verified.
NVIDIA has released Kumo Tabular, an open-source tabular foundation model that predicts labels for new rows in a single forward pass from labeled table rows, supporting classification and regression without training, tuning, or feature engineering. The model comes in three sizes from 28M to 215M parameters, was pretrained entirely on synthetic tables, is commercially usable under the OpenMDW-1.1 license, and ranks first on four benchmarks: TabArena, BeyondArena, TALENT, and ScoringBench.
Why it matters: The source reports model sizes, training-data provenance, and rankings on four benchmarks, helping readers assess the accuracy and efficiency tradeoffs of tabular foundation models.
Microsoft Research has released Quine, an AI research system designed for biological complexity. It combines a biological world model with an interactive harness connecting models, scientific tools, literature, and experimentalists. In work with the Broad Institute on pancreatic ductal adenocarcinoma, it predicted and prioritized thousands of compounds; the top-ranked compound produced the largest classical-to-ground-state transition in wet-lab experiments. The process from narrowing the search space to selecting a candidate for validation took one weekend.
Why it matters: Microsoft Research describes Quine's multimodal biological world model and experimental loop, including wet-lab validation in pancreatic cancer compound screening.
Xiaomi has released the MiMo-V2.6 family, including the multimodal MiMo-V2.6-Pro and MiMo-V2.6-Flash, and introduced MiMo-V2.6-Pro-UltraSpeed, which delivers up to 20× faster output.
Why it matters: Xiaomi's first entry in the frontier open-model lineup, with open RL environments and training recipes, provides a concrete example of post-training cost strategies.
Researchers from @BroadInstitute, @UniofExeter, and other institutions are using AlphaGenome Atlas to better identify potentially disease-causing DNA variants and interpret their effects. 🧵
Weather forecasting plays a key role in the transition to renewable energy. WeatherNext 3 provides grid operators and wind and solar power companies with forecasts of wind speeds at turbine height and solar irradiance, supporting smarter, cleaner energy planning. 🧵
Google DeepMind and Google Research have released WeatherNext 3, which they call the most advanced and accurate global weather model so far, based on independent real-time evaluations by Brightband.
Why it matters: The article reports specific measures for resolution, update frequency, and precipitation accuracy, helping readers assess AI weather forecasting on real-time data.
Google has introduced TimesFM-3, a time-series foundation model pretrained natively for multivariate forecasting. It has 330M parameters, was pretrained on more than one trillion time points, and forecasts in a single forward pass. It supports multiple targets, historical covariates, and known future covariates, and ranks first for both point and probabilistic forecasting on the three public benchmarks Gift-Eval, FEV-Bench, and Time.
Microsoft has released GigaPath-Flash and GigaTIME-Flash, using a 22M-parameter ViT-S backbone distilled from the billion-parameter GigaPath encoder to sharply reduce the compute requirements of pathology foundation models. Both are open-source under Apache 2.0.
IBM's Granite team has released Granite 4.2, its first family of dense decoder-only reasoning models, in 3B, 8B, and 30B sizes. They were post-trained from the Granite-4.1 base and are all licensed under Apache 2.0.
Why it matters: The article discloses Granite 4.2's architecture, SFT data mix, and multi-stage RL curriculum in detail, offering a look at concrete post-training methods for reasoning models.
Hugging Face's OlmoEarth Studio can now compute and export embeddings; its code and model weights are public. Through the UI or API, users can select a region, time range, encoder variant (Nano: 128 dimensions and 1.4M parameters; Tiny: 192 dimensions and 6.2M parameters; Base: 768 dimensions and 89M parameters), resolution, and imagery source, then export an int8 Cloud-Optimized GeoTIFF.
Google DeepMind has released WeatherNext, an AI model that achieves state-of-the-art forecasts of cyclone tracks, intensity, and wind-field structure, providing about one additional day of lead time on average—roughly a decade of progress in meteorology.
Why it matters: The source reports measured gains of about one additional day of warning for cyclone tracks and intensity and opens the model weights and code.
Google has released TabFM, a zero-shot foundation model for classification and regression on tabular data. It recasts table prediction as an in-context-learning problem, producing predictions in a single forward pass without manual training, tuning, or feature engineering.
Why it matters: The source details TabFM's architecture and TabArena results, helping readers assess whether zero-shot table prediction can replace traditional tree-model workflows.
Google Research has open-sourced its hydrology framework on GitHub under Apache 2.0, enabling national weather and hydrology agencies to train AI flood-forecasting models using the same architecture and similar training data as Google Flood Hub.
