Total 671 · Clear filters
On 17 March 2025, Chinese AI company 01.AI released Wanzhi, an enterprise one-stop platform for large models that supports DeepSeek, Qwen and its own Yi models, alongside customized DeepSeek deployment and fine-tuning tools.
On 17 March 2025 Mistral AI released Mistral Small 3.1, an Apache 2.0 model that adds image understanding and a 128K context window while running on a single RTX 4090 or a 32GB Mac.
On 16 March 2025 Baidu launched ERNIE X1, a deep-thinking reasoning model and its first multimodal reasoning model able to use tools. Baidu said it matched DeepSeek-R1 at half the price.
On 16 March 2025 Baidu released ERNIE 4.5, a native multimodal foundation model, together with the reasoning model ERNIE X1, and made ERNIE Bot free to individual users ahead of schedule.
On 13 March 2025, Ai2 released OLMo 2 32B, which it called the first fully open model to beat GPT-3.5-Turbo and GPT-4o mini on a suite of academic benchmarks, and published everything needed to reproduce it.
On 13 March 2025 Cohere introduced Command A, a 111-billion-parameter model that runs on two GPUs. Cohere claimed it matches or beats GPT-4o and DeepSeek-V3 on agentic enterprise tasks.
On 12 March 2025 Google released Gemma 3, a family of four model sizes from 1B to 27B with image input and a 128,000-token context window, alongside ShieldGemma 2, an image-safety checker.
On 12 March 2025, Google DeepMind introduced two robotics models built on Gemini 2.0: Gemini Robotics, a vision-language-action model that outputs robot actions, and Gemini Robotics-ER, which focuses on spatial and embodied reasoning.
On 11 March 2025 OpenAI released the Responses API and an open-source Agents SDK, merging the simplicity of Chat Completions with the tool use of the Assistants API and bundling web search, file search, computer use and tracing.
On 10 March 2025 AgiBot released GO-1, a general-purpose embodied foundation model the company says can run across several of its humanoid robot bodies, learn from human videos and pick up new tasks from a few examples.
On 6 March 2025 the Qwen team released QwQ-32B, a 32-billion-parameter reasoning model that it said matches DeepSeek-R1, which has 671B parameters, on its benchmarks. It is available with open weights.
On 6 March 2025 Mistral AI launched Mistral OCR, a document-understanding API that it said processes up to 2,000 pages per minute, handles thousands of scripts and languages, and scored above GPT-4o on its benchmark.
On 4 March 2025 Cohere's research lab released Aya Vision in 8B and 32B open-weight sizes. The models understand images and text across 23 languages and carry a non-commercial licence.
Huawei's Atlas 900 A3 SuperPoD, launched in March 2025, links up to 384 Ascend 910C chips into one logical computer rated at up to 300 PFLOPS, and is sold in cloud form as CloudMatrix384.
In March 2025 AI21 Labs released Jamba 1.6, a model family packaged for private enterprise deployment with weights under the Jamba Open Model License, and launched Maestro, a planning and orchestration system.
In February 2025, Ant Group's InclusionAI released the Ling (Bailing) model family, with Ling-Lite at 16.8B total parameters and Ling-Plus at 290B, weights under the MIT license, and, the team said, training without premium GPUs.
OpenAI released GPT-4.5, code-named Orion, on 27 February 2025 as an expensive research preview that was later shut down.
On 27 February 2025 Tencent released Hunyuan Turbo S, saying it replies faster than DeepSeek-R1. The model was offered through a Tencent Cloud API, and its weights were not released.
On 26 February 2025 Amazon introduced Alexa+, a generative-AI assistant priced at $19.99 a month and free for Prime members. Anthropic announced the same day that Claude was part of Alexa+.
On 26 February 2025 Microsoft announced Phi-4-mini and Phi-4-multimodal, both released under the MIT licence on Hugging Face and available the same day in Azure AI Foundry and GitHub Models.
On 25 February 2025 Alibaba open-sourced the Wan2.1 video generation models under Apache 2.0, with inference code and weights. The release included text-to-video and image-to-video models.
