New
Everything new in AI as it appears: launches announced by the labs, brand-new models, datasets, papers and open-source tools. Checked every minute or two. Launches feed · Trending · News
Live now
- Connecting…
Yesterday
- Datasetctjones/psychosis-rl-generations
Psychosis RL teacher and baseline generations Paired synthetic responses for studying the effects of an RL adapter on Qwen/Qwen3-30B-A3B. finetuned uses CDBiddulph/qwen3-
- DatasetOneWorld-AI/AgiBotWorld-Beta-processed
AgiBot World Beta processed episodes A processed subset of agibot-world/AgiBotWorld-Beta, containing 142,685 verified episodes, approximately 2,330.1 hours, and 6.22 TB o
- Datasetctjones/wordchain-rl-generations
Wordchain RL teacher generations Reasoning and final-answer samples generated by Qwen/Qwen3-30B-A3B with the CDBiddulph/qwen3-30b-a3b-wordchain-rl adapter. These are gene
- DatasetIO-AIR/filterme-sft
filterme-sft · v1.7 Current release: v1.7 — the repo root always holds the latest release: finals (train.jsonl.gz, manifest.jsonl.gz), per-source provenance pools (shards
- Dataset
- DatasetGunGG4/paiec-Qwen8Binstruct-guneesh-data
PAIEC: Qwen3-VL-8B-Instruct item embeddings and IRT fits Training artifacts made for the Stanford AIMS Predictive AI Evaluation Competition (PAIEC), NeurIPS 2026. Everyth
- Dataset
- Dataset
- Datasetsheepy928/digital-maxima-eval
Digital MAXIMA evaluation data (gated) Access is approved by hand. This dataset holds the hidden-split manifests of the maxima-twin-bench benchmark. This page and the fil
- Datasetvovaxa44/snapdragon778g-llama-cpp-opencl-bench
Snapdragon 778G (Adreno 642L) — llama.cpp CPU vs OpenCL benchmark Measured throughput of Qwen2.5-3B-Instruct Q4_K_M in llama.cpp on an OPPO Reno7 5G (Snapdragon 778G, Adr
- Dataset
- Datasetautobio/zymedb-reaction50-splits-v1
ZymeDB reaction-cluster split (version 1) Methods We reused the release's rxn_cluster_50 assignments: upstream DRFP fingerprints clustered by RDKit Butina using Tanimoto
- DatasetJerry999/RSIChallenge
RSIChallenge — artifacts / 工件集 Composed on the aws server. Source repos: auto-impl (PostTrainBench agent study) and RSIChallenge (recursive self-improvement study). Built
- DatasetRESMP-DEV/mimo-rl-telemetry-archive
MiMo RL telemetry archive This dataset preserves the public Xiaomi MiMo RL training telemetry feed from https://mimo.xiaomi.com/rl/. It contains aggregate training teleme
- Dataset
- Datasetanonymous-7989/herikgqa-benchmark
HeriKGQA: benchmark and dataset snapshot A 500-question benchmark for question answering over a cultural-heritage sensor knowledge graph, and the anonymised dataset snaps
- DatasetBeifengdo/DMU-Postgraduate-Interview-and-Transfer
version https://git-lfs.github.com/spec/v1 oid sha256:e53e8a8bb26672b91a1d4a05a195503a3f30e702332687097ae0082b846c56e8 size 481
- Dataset
- Dataset
- Datasetxmarva/tfs-tatoeba-en-ru
Tatoeba English–Russian, cleaned and split without leaks Built in lesson 1 of the course "Transformers: from scratch to production": https://github.com/xmarva/transformer
- DatasetBeifengdo/Marine-Measurement-Control-Computer-Software-Fundamentals
version https://git-lfs.github.com/spec/v1 oid sha256:986ad9331bed026bbff422287ffe1e65e8841940c696b9557531455320f7c7df size 496
- DatasetBeifengdo/Marine-Measurement-Control-Circuit-Labs
version https://git-lfs.github.com/spec/v1 oid sha256:1e77fa9eb725406c17694d92b7e0c8f8e70934cf58bee706b018d6d29ce37a37 size 471
- DatasetBeifengdo/Marine-Measurement-Control-Circuit-Principles
version https://git-lfs.github.com/spec/v1 oid sha256:a4288ab5be87abb5f0c7a00e9d8606331fbe0ff29bb53afbe549465920ea02f5 size 471
- Dataset
- DatasetBeifengdo/Marine-Measurement-Control-Sensor-Labs
version https://git-lfs.github.com/spec/v1 oid sha256:b2c5b4ce34b2cf21fed21603861c9ef77a89cfc4c8f93ae858259e7a9458f8e5 size 466
- Datasethbfreed/tiny-dark-stories
Tiny Dark Stories Every story in TinyStories (V2), paired with a gently spooky twin written from the same character, setting and a strange little premise. 2,699,372 train
- Dataset
- Dataset
- Dataset48karem/turkish_50_question_for_answer_dataset.txt
this dataset can be used and modified but don`t use for commercial activities if you use or modify this dataset, please provide appropriate attribution to 48karem (kerem
- DatasetBeifengdo/Ship-Engineering-Major-Materials
version https://git-lfs.github.com/spec/v1 oid sha256:f0fa2cbc805a4ec1aae82bf020776a341536acf820c6c9c24fd3ffa4d07e0341 size 450
- DatasetBeifengdo/Principles-of-Naval-Architecture
version https://git-lfs.github.com/spec/v1 oid sha256:ff1cb3a4466b88ab21681460892eb635e298738451502bbf2c9efae526d351cb size 447
- DatasetBeifengdo/Ship-Hydrostatics
version https://git-lfs.github.com/spec/v1 oid sha256:3e6cf5e6336b7d4c4ebab114af62fdf064c05133405ea1324d48b82c269651af size 432
- DatasetOneWorld-AI/RoboMIND2.0-processed
RoboMIND 2.0 processed episodes A processed subset of RoboMIND 2.0 containing 134,537 episodes and approximately 846 hours of bimanual robot demonstrations. This is a der
- DatasetBeifengdo/Marine-Electrical-and-Electronic-Engineering
version https://git-lfs.github.com/spec/v1 oid sha256:69a41ea043c5406c2d78ddfa9d31de0ab5e926bc1933adf6ef9ad98e7cac1f7e size 465
- DatasetBeifengdo/Ship-Safety-Training
version https://git-lfs.github.com/spec/v1 oid sha256:e732e34737a3a092bdf48c677d4b305a73f534ebd6fd7d9d4f3ec311de31a929 size 440
- Datasetabhijayrana/erp-bench
ERP-Bench ERP-Bench is the Odoo 19 benchmark used in the paper "Anchor: Preventing Artifact Drift for Reliable Agent Verification." It contains 300 long-horizon procureme
- DatasetBeifengdo/Sampling-Surveys
version https://git-lfs.github.com/spec/v1 oid sha256:bc9bc029057426de860971c51351f89337034b57b36342b5b76a3132b254a0d8 size 428
- DatasetBeifengdo/Testing-Technology
version https://git-lfs.github.com/spec/v1 oid sha256:449ebe51659242690beb9166e184af2f63002075601d39f1de226df285f8559c size 430
- Datasetmchen04/coverage-map-data
Coverage Map demo CTs Two public, deidentified CT series from The Cancer Imaging Archive (TCIA), served as plain DICOM files for the Coverage Map demo viewer. Research us
- Dataset
Headlines and links come from the labs' own blogs, Hacker News and the GDELT news index, and from the sources our index already follows. We show a short excerpt and always link to the original. A launch reported by a news site rather than the lab is marked “Reported” until the lab itself announces it.