Cohere releases Command A Vision
On 31 July 2025 Cohere released Command A Vision, a 112-billion-parameter vision-language model built on Command A. Its open weights were posted under a non-commercial licence.
Total 671 · Clear filters
Linux Foundation welcomes the AGNTCY project
On 29 July 2025 the Linux Foundation welcomed AGNTCY, an open project for interoperable AI agents, with Cisco, Dell Technologies, Google Cloud, Oracle and Red Hat as formative members.
Alibaba open-sources Wan2.2, a mixture-of-experts video model
On 28 July 2025 Alibaba released Wan2.2, a video generation model family, as open code and weights under Apache 2.0. Its A14B models brought a mixture-of-experts design to video diffusion.
Tencent open-sources HunyuanWorld 1.0 for 3D world generation
On 26 July 2025 Tencent open-sourced HunyuanWorld 1.0, a model that builds explorable 3D worlds from text or images, with a technical report.
Unitree releases the R1 humanoid
On 25 July 2025 Unitree released the R1, a roughly 25 kg humanoid robot with 26 joints, priced from 39,900 yuan — well below the 150,000 to 200,000 yuan prices then common for humanoid robots.
StepFun releases the open Step 3 multimodal reasoning model
In July 2025, during WAIC, StepFun released the open Step 3 multimodal reasoning model and formed a Model-Chip Ecosystem Innovation Alliance with Huawei, Biren, Moore Threads and Enflame.
Runway introduces Aleph video editing model
Runway released Aleph, an in-context video model that edits footage already shot: it adds, removes and transforms objects, generates new camera angles and changes style or lighting, and is available on all paid plans.
StepFun releases Step-Audio 2
On 23 July 2025 StepFun released the technical report for Step-Audio 2 along with two new benchmarks. Open mini versions followed on 29 August 2025.
Infinigence AI publishes Megrez2 with cross-layer expert sharing
Infinigence AI released Megrez2-Preview on 22 July 2025 and the finished Megrez2-3x7B-A3B in September, an on-device mixture-of-experts model in which three adjacent layers share one pool of 64 experts.
Alibaba releases Qwen3-Coder and the Qwen Code agent
On 22 July 2025 Alibaba announced Qwen3-Coder-480B-A35B-Instruct and released Qwen Code, an open-source command-line agent adapted from Gemini CLI.
NAVER releases a pruned, distilled reasoning model in the HyperCLOVA X SEED line
On 21 July 2025 NAVER published HyperCLOVA X SEED 14B Think, a reasoning model built on top of a pruned and distilled model, with training reported at 68,049 A100-80GB GPU hours.
Lovable raises a $200M Series A at a $1.8B valuation
On 17 July 2025, Lovable announced a $200M Series A led by Accel at a $1.8B valuation, eight months after launch. It calls the round one of Europe's largest, and says its mission is letting people who cannot code build software by talking to AI.
OpenAI launches ChatGPT agent
OpenAI's ChatGPT agent, launched on 17 July 2025, carries out multi-step tasks on its own virtual computer — browsing websites, running code and producing editable files — combining Operator's web interaction with deep research's ability to gather and summarise information.
LG AI Research releases EXAONE 4.0 with a reasoning mode
LG AI Research published EXAONE 4.0 on 15 July 2025, a two-size family with a switchable reasoning mode. The 1.2B model targets on-device use, and a licence change allows commercial deployment through partners.
Mistral AI releases Voxtral speech models
On 15 July 2025 Mistral AI released Voxtral Small and Voxtral Mini, two Apache-licensed speech models that transcribe up to 30 minutes of audio, answer questions about recordings and produce structured summaries.
Cognition acquires Windsurf
In July 2025 Cognition said it had signed a definitive agreement to acquire Windsurf, the agentic IDE, including its IP, brand, people and $82M of annual recurring revenue.
Perplexity launches the Comet AI browser
On 9 July 2025 Perplexity released Comet, a Chromium-based browser whose assistant can carry out multistep tasks such as booking meetings or shopping. Access began with Max subscribers and an invite list, then widened to a free download and mobile apps.
xAI releases Grok 4 and Grok 4 Heavy
On 9 July 2025 xAI released Grok 4 and Grok 4 Heavy. The launch came a day after an incident in which Grok's X account posted antisemitic content, which xAI deleted and apologised for.
Hugging Face releases SmolLM3, a small reasoning model
On 8 July 2025 Hugging Face released SmolLM3, a 3B-parameter open model that handles up to 128K tokens of context and switches between reasoning and direct-answer modes. The full training recipe was published with it.
Zhipu releases GLM-4.5 for agentic tasks under the MIT License
In July 2025 Zhipu released GLM-4.5 and GLM-4.5-Air under the MIT License. Both are hybrid reasoning models built for agentic tasks, and GLM-4.5 ranked third in a 12-benchmark comparison.
Moonshot AI open-sources the trillion-parameter Kimi K2
In July 2025 Moonshot AI released Kimi K2 as an open-weight mixture-of-experts model with 1 trillion total parameters and 32 billion active per token, under a modified MIT license.