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…
Thursday 22 October
- DatasetCalamity228/GPT-6-Luna-RU
GPT-6 Luna General Purpose (RU) Синтетический русскоязычный диалоговый датасет общего назначения (General Purpose) для supervised fine-tuning языковых моделей. Все 6 015
- Datasetjerryyan/routerbench-trajectories
routerbench trajectories Full agent trajectories and graded results for routerbench, a study of task-level model routing for agentic coding. Code, explanations, and curre
- DatasetDsxsaim/fatima-fellowship-eval
Evaluating AI's Over-Normalization Blind Spot Question 1: The Blind Spot I have experienced this a lot actually. Working as a Data & Systems Associate in a hospital, I no
- Dataset
- Datasetvdags-lr/test_multi
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names":
- Dataset
- Dataset
- Datasetvdags-lr/test_multi_20261022_230922
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names":
- Dataset
- Dataset
- DatasetPTeterwak/openwebrl-arm-matched-pairs
OpenWebRL ARM preference pairs, matched action counts These are preference pairs for fine-tuning a browser agent (OpenWebRL-4B-SFT, Qwen3-VL-4B) with DPO or RPO. Each row
- DatasetSTT-DATASET/en-accent-zaebuc-ar-en-noncommercial
zaebuc_ar_en_NONCOMMERCIAL ASR-ready version of ZAEBUC-Spoken (NYU Abu Dhabi) (license: CC-BY-NC-SA-4.0 (NON-COMMERCIAL)). Arabic-English code-switched Zoom role-plays, U
- DatasetSTT-DATASET/en-accent-voxpopuli-en-accented
voxpopuli_en_accented ASR-ready version of facebook/voxpopuli (en_accented) (license: CC0). European Parliament speeches by non-native speakers of 15 L1s (Dutch, German,
- DatasetTomasAnderegg/so101_pill_sorting_v3
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names":
- Dataset
- DatasetSTT-DATASET/en-accent-vctk-accents
vctk_accents ASR-ready version of VCTK 0.92 (Edinburgh DataShare) (license: CC-BY-4.0). 110 speakers reading newspaper sentences; accent column = VCTK accent (English, Sc
- Datasetzinuzoid/record-test_20261022_220126
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names":
- Datasetavaliev/oracle-audit-amp-scores
Data card — Oracle Audit: per-peptide scores from 16 in-silico AMP oracles Files file shape description oracle_audit_scores_wide.csv 14,977 rows x 20 cols one row per pep
- DatasetSTT-DATASET/en-accent-speechocean762-mandarin-l1
speechocean762_mandarin_l1 ASR-ready version of mispeech/speechocean762 (license: CC-BY-4.0). Mandarin-L1 English read speech, half children (see meta.age / meta.child).
- DatasetSTT-DATASET/en-accent-slr83-uk-ireland
slr83_uk_ireland ASR-ready version of OpenSLR 83 (license: CC-BY-SA-4.0). UK & Ireland regional English (accent column = dialect). Format: parquet, 500-row row groups (Co
- DatasetLayerLens/hubspot-open-deal-count-test
HubSpot Open-Deal Count Traces LayerLens | Blog post | Analysis script | Notebook 2 agents, built on Claude Sonnet 5.5 and Claude Sonnet 5, were asked how many deals are
- DatasetSTT-DATASET/en-accent-slr70-nigerian
slr70_nigerian ASR-ready version of OpenSLR 70 (license: CC-BY-SA-4.0). Nigerian English read speech (Lagos, London). Format: parquet, 500-row row groups (Cohere training
- DatasetSTT-DATASET/en-accent-silencio-accents-noncommercial
silencio_accents_NONCOMMERCIAL ASR-ready version of SilencioNetwork/english-accents-speech (license: CC-BY-NC-4.0 (NON-COMMERCIAL)). Read monologues from 35 countries (Ph
- DatasetSTT-DATASET/en-accent-nchlt-south-african
nchlt_south_african ASR-ready version of NCHLT English Speech Corpus (SADiLaR) (license: CC-BY-3.0). South African English, short read prompts, smartphone recordings. Sou
- Datasetkushasareen/held-out-math-eval
held-out-math-eval Four held-out math sets, as parquet files in the format the verl validation loop reads. Every row asks for the answer in \boxed{} with the same system
- DatasetSTT-DATASET/en-accent-nchlt-aux-south-african
nchlt_aux_south_african ASR-ready version of NCHLT English auxiliary 1+2 (SADiLaR) (license: CC-BY-3.0). Additional South African English read prompts. Format: parquet, 5
- DatasetSTT-DATASET/en-accent-mixat-emirati-ar-en-noncommercial
mixat_emirati_ar_en_NONCOMMERCIAL ASR-ready version of MBZUAI/MIXAT (license: no license stated (research use) - NON-COMMERCIAL). Emirati Arabic-English code-switched spe
- DatasetSTT-DATASET/en-accent-lwazi-south-african-8k
lwazi_south_african_8k ASR-ready version of Lwazi English ASR corpus (SADiLaR) (license: CC-BY-2.5-ZA). South African English telephone speech (8 kHz origin, upsampled to
- DatasetSTT-DATASET/en-accent-librivox-australian-verified
librivox_australian_verified ASR-ready version of ablmontazer/australian-english-speech (LibriVox readers of Australian books) (license: CC0-1.0). Only readers that the C
- Dataset
- Dataset
- DatasetSTT-DATASET/en-accent-imda-singapore-subset
imda_singapore_subset ASR-ready version of IMDA National Speech Corpus via mesolitica/IMDA-STT (license: Singapore Open Data Licence v1.0). Speaker-balanced ~300 h subset
- Datasetzinuzoid/record-test_20261022_220002
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names":
- Datasetkushasareen/openmath-hard-64
openmath-hard-64 64 competition-style math problems with integer answers, taken from nvidia/OpenMathReasoning (CC-BY-4.0), split cot. How the problems were chosen From th
- DatasetSTT-DATASET/en-accent-globe-v2
globe_v2 ASR-ready version of MushanW/GLOBE_V2 (license: CC0-1.0). Common Voice 14 English, enhanced. ~69% US English; filter on accent for non-US/UK speakers. Overlaps c
- DatasetSTT-DATASET/en-accent-escwa-ar-en-noncommercial
escwa_ar_en_NONCOMMERCIAL ASR-ready version of arbml/ESCWA (license: no license stated - NON-COMMERCIAL). UN ESCWA meetings, Arabic-English code-switching, cut on the sou
- DatasetSTT-DATASET/en-accent-egyptian-ar-en-cs-noncommercial
egyptian_ar_en_cs_NONCOMMERCIAL ASR-ready version of MohamedRashad/arabic-english-code-switching (license: tagged GPL, YouTube-sourced - NON-COMMERCIAL / legal risk). Egy
- DatasetSTT-DATASET/en-accent-edacc
edacc ASR-ready version of edinburghcstr/edacc (license: CC-BY-SA-4.0). Edinburgh International Accents of English: conversational Zoom speech, ~40 L1/accents. Source val
- DatasetSTT-DATASET/en-accent-earnings22-cut-v2
earnings22_cut_v2 ASR-ready version of Earnings-22 (revdotcom) full calls (license: CC-BY-SA-4.0). Earnings calls from 27 countries, re-cut with the new pipeline: Soniox
- DatasetSTT-DATASET/en-accent-common-accent
common_accent ASR-ready version of DTU54DL/common-accent (license: MIT). Non-US/UK subset of Common Voice: India/South Asia, German, South African, Filipino, Singaporean,
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.