Baidu releases the open-weight ERNIE 5.5 Flash Base checkpoint
On 5 November 2026 Baidu introduced ERNIE 5.5 Flash Base, an open-weight 280B-parameter mixture-of-experts pre-training checkpoint published on Hugging Face under Baidu's own license.
Total 374 · Clear filters
NVIDIA releases Nemotron 3.5 Transcribe and Granary 2
On 28 October 2026 Nvidia released Nemotron 3.5 Transcribe, an open-weight, free speech recognition model covering up to 40 languages, alongside the second version of its Granary speech dataset.
PFN releases a 53-language translation model built with NICT
On 26 October 2026 PFN released the weights of PLaMo 3 Translate, a 53-language translation model built on a base model from joint research with NICT, in four sizes from 0.8B to 31B and several quantised formats.
The Agentic AI Foundation passes 300 members in under a year
The Agentic AI Foundation reported 305 member organisations in under a year, saying agents are moving from developer experiments to institutional infrastructure.
AAIF adds Agent Skills, Open Responses, OpenEnv and OpenSandbox
On 22 October 2026 the Agentic AI Foundation added four projects — Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba — bringing it to 10 hosted projects covering different layers of the agentic stack.
Anthropic donates Agent Skills to the Agentic AI Foundation
On 22 October 2026 Anthropic handed the Agent Skills open standard — its name, logo, GitHub organisation and domain — to the Agentic AI Foundation, a Linux Foundation project, after more than 70 agent platforms had adopted it.
Liquid AI adds an omni-modal model to the LFM3 family
On 21 October 2026 Liquid AI released LFM3-Omni-2B, the first omni-modal model in its LFM3 family, taking text, images, video and audio as input and returning text or speech. At 2.17B parameters it targets phones and laptops.
DeepSeek updates V4.1-Flash and opens its Search and Fetch APIs
On 17 October 2026 DeepSeek shipped DeepSeek-V4.1-Flash-1017, an update to its smallest model with better benchmarks, faster task completion and shorter outputs; its weights were published and its Search and Fetch APIs opened at $0.75 per 1,000 calls.
IFM and Inception42 release the six-model Jais 3 family
On 15 October 2026, IFM and Inception42 announced Jais 3, six Arabic-centric language models from 900 million to 375 billion parameters plus a speech recogniser, evaluated on 24 audited Arabic benchmarks.
Kuaishou rolls out Kling 4.0
On 15 October 2026 Kuaishou's video generation model Kling 4.0 became available to all users, adding longer shots, more dynamic motion and Omni Reference, with a faster Flash variant released alongside it.
Meta releases the open-weight Muse Spark Open
On 15 October 2026 Meta released Muse Spark Open, an open-weight, natively multimodal model with 816 billion total parameters, optimised to run on Apple's Mac Studio. Its weights were published on Hugging Face.
Mistral AI unveils Mistral Large 4, a trillion-parameter open-weight model
On 6 October 2026 Mistral AI opened a public preview of Mistral Large 4, a model with about 1 trillion total parameters. On 29 October 2026 it released open weights of an improved version.
Aleph Alpha releases the Apache-2.0 model Kolibri
On 3 October 2026 Aleph Alpha released Kolibri, an open-weight model pitched at public administration, industry and aerospace, with German-language specialization and a traceable training process presented as its advantage rather than raw scale .
MiniMax releases MiniMax-M3.1-Flash
MiniMax posted the weights of M3.1-Flash to Hugging Face on 17 October 2026. The model is a multimodal M3-line release with about 428 billion parameters, MiniMax Sparse Attention throughout, and DSpark speculative decoding.
Moonshot AI releases Kimi K3.1 Pro
In October 2026 Moonshot AI released Kimi K3.1 Pro, an open-weights upgrade for financial research, 3D modeling and coding. It was the first model in the K3.1 series.
ElevenLabs launches Eleven v4, its most expressive text-to-speech model
On 28 September 2026 ElevenLabs released Eleven v4, a text-to-speech model it calls its most emotive yet, alongside a low-latency Turbo variant; the company says it topped Artificial Analysis and was preferred by about 75% of listeners in blind tests.
PixVerse releases R2 and turns its world model into playable games
On 22 September 2026 PixVerse announced R2, a scaling release of its real-time world model that adds more data, tasks and input types at the same latency, and showed it running playable, prompt-editable worlds.
Character.ai post-trains its own CAI-Image models on Qwen-Image
Character.ai's CAI-Image family consists of image models post-trained from open-source Qwen-Image and tuned for character-driven storytelling, so the same character stays recognisable across styles, scenes and pages. The company announced the models on 16 September 2026.
DeepSeek releases DeepSeek-V4.1-Flash with a causal encoder–decoder design
On 10 September 2026 DeepSeek released DeepSeek-V4.1-Flash, a multimodal model with a causal encoder-decoder design that splits its active parameters between reading input and writing output. Its KV cache is much smaller than that of the previous generation.
Suno releases v6, built with Warner Music Group, BMG and Believe
On 9 September 2026 Suno released v6, a new generation of music models built with Warner Music Group, BMG and Believe, offered in three versions and paired with safeguards that screen uploaded audio and lyrics.
OpenBMB publishes MiniCPM5-2B
On 6 September 2026 OpenBMB released MiniCPM5-2B, a dense 2-billion-parameter open model that the team describes as state of the art for its size class, released together with the UltraX training dataset.