Total 671 · Clear filters
On 21 November 2024, Ai2 released Tülu 3, two models (8B and 70B) fine-tuned from Llama 3.1 with a fully published post-training recipe that adds reinforcement learning with verifiable rewards to SFT and preference tuning.
On 18 November 2024 Mistral AI released Pixtral Large, a 124-billion-parameter open-weight multimodal model built on Mistral Large 2, alongside Mistral Large 24.11 and a Le Chat update adding search, Canvas, image generation and agents.
On 6 November 2024 XPENG showed IRON, a humanoid robot, at its AI Day in Guangzhou, saying it shares technology with the company's AI vehicles and is already doing trial work in its own car factory.
On 5 November 2024 Tencent open-sourced Hunyuan-Large, a 389-billion-parameter mixture-of-experts model, one day after submitting its technical report to arXiv.
On 4 November 2024 Anthropic made Claude 3.5 Haiku available at four times the price of Claude 3 Haiku, saying the new model was more intelligent. About a month later the company lowered its price.
In November 2024 Hugging Face released SmolLM2, a family of small language models at 135M, 360M and 1.7B parameters, with open weights, an instruct variant and a technical report that made the recipe public.
On 31 October 2024, Physical Intelligence announced π0, a prototype robot-control model trained on large-scale multi-task and multi-robot data that can be prompted or fine-tuned to control a variety of robots.
On 24 October 2024 iFlytek released Spark 4.0 Turbo at its Global 1024 Developer Festival, claiming gains over GPT-4 Turbo and GPT-4o, and announced a jointly built computing platform and plans for a reasoning model.
On 22 October 2024 Anthropic announced an upgraded Claude 3.5 Sonnet, a new Claude 3.5 Haiku and computer use in public beta, which lets Claude operate a computer through screenshots, cursor movements and typing.
On 22 October 2024 Runway released Act-One inside Gen-3 Alpha, a tool that turns footage of an actor's performance and voice into an animated character without motion capture or manual face rigging.
On 22 October 2024 Stability AI released Stable Diffusion 3.5 Large and Large Turbo, followed by the smaller Medium model on 29 October, all under a community license that is free for commercial use below $1 million in annual revenue.
In October 2024 01.AI launched Yi-Lightning, which placed sixth on the LMSYS Chatbot Arena and first among Chinese models, according to Kai-Fu Lee. Its API was priced at 0.99 yuan per million tokens.
On 16 October 2024 Mistral AI released Ministral 3B and 8B, small models built for on-device and edge use, with 128K-token context, low per-token API prices and research access to the 8B Instruct weights.
On 4 October 2024 Meta premiered Movie Gen, a set of media foundation models that generate video, images and sound. Meta did not release the models, saying it would first work with filmmakers and creators.
On 2 October 2024 Black Forest Labs released FLUX1.1 [pro], a faster version of its text-to-image model, and made its BFL API generally available at 4 cents per image.
In October 2024 Cohere's research lab released Aya Expanse in 8B and 32B sizes. The models cover 23 languages, including Japanese, and were compared with several rival models in judged win-rate tests.
On 25 September 2024 Ai2 released Molmo, a family of open vision-language models ranging from MolmoE-1B to Molmo-72B that can point at things in images; Ai2 said Molmo-72B ranked just behind GPT-4o in human preference.
On 25 September 2024 Meta released Llama 3.2, which added image understanding to the Llama family with 11B and 90B vision models and added small 1B and 3B text models for phones and other edge devices.
On 19 September 2024 Alibaba released Qwen2.5 with 13 open-weight models. The family was trained on up to 18 trillion tokens and supports 128K tokens of context in more than 29 languages.
On 17 September 2024 Mistral AI released Pixtral 12B, an open-weight model pairing a 400M vision encoder with a 12B decoder and scoring 52.5% on MMMU and 90.7 on DocVQA.
On 12 September 2024 OpenAI introduced o1-preview and o1-mini, its first reasoning models, designed to think longer before responding. The full o1 followed in December 2024, and Microsoft made the preview models available on Azure the same day.
