Ling 3.0 Flash is the smartest open model in its size class. On the Artificial Analysis Intelligence Index, Ling 3.0 Flash scores 38 points, a big jump over its predecessor. That puts it on par with Qwen3.6 27B while using far fewer active parameters. It still trails the leader, DeepSeek V4 Flash, which sits at 52 points. According to Artificial Analysis, it's the smartest open model under 124 billion total parameters. No smaller model matches its score.
Image: Artificial Analysis
On the AA Omniscience test, the hallucination rate dropped from 97 to 44 percent compared to the previous version. The model now refuses to answer questions it doesn't have reliable answers for far more often. Ling 3.0 Flash also shows strong gains in agentic tasks over its predecessor, including on the t3-Bench Banking benchmark.
On per-token pricing, Ling 3.0 Flash beats every comparably capable model. It does burn through more tokens on complex tasks than similarly strong alternatives. But even on a per-task basis, it stays cheaper than Qwen3.6 27B. Ant Group's inclusionAI is releasing the model under an MIT license on the inclusionAI API and through DeepInfra. Model weights are available directly on Hugging Face.
The Trump administration discussed sanctions and cloud bans targeting Chinese open-weight AI models, according to the New York Times. OpenAI and Anthropic pushed for restrictions, while Nvidia, Google, and Meta fought back. After pushback from Silicon Valley, Washington backed off for now, but a decision is expected before Xi Jinping’s visit in September.
According to benchmark platform Artificial Analysis, Nvidia's new Nemotron 3 Ultra is the most capable open AI model from the US to date. It has roughly 550 billion total parameters, with about 55 billion active at any given time. On the Artificial Analysis intelligence ranking, Nemotron 3 Ultra scores 48 points, well ahead of other open US models like Gemma 4 31B (39), Nemotron 3 Super (36), and gpt-oss-120b (33). It doesn't reach the top open models from China, though. Kimi K2.6 scores 54 points there. The current strongest closed model, Opus 4.8, hits 61 points.
Nemotron 3 Ultra lands in the "most attractive quadrant" of the Artificial Analysis chart, combining high intelligence scores with fast output speed. | Image: AAII
On provider DeepInfra, Nemotron 3 Ultra also delivers more than 300 tokens per second, according to Artificial Analysis. Comparably sized models from DeepSeek or Moonshot currently manage only 50 to 100. Nvidia says the model will be released on June 4 on Hugging Face, OpenRouter, and other platforms.
Canadian AI company Cohere is releasing its most powerful language model, Command A+, as open source under the Apache 2.0 license. The mixture-of-experts model has 218 billion parameters with 25 billion active, and already runs on two Nvidia H100 GPUs or a single Blackwell GPU, according to Cohere.
Command A+ is built for enterprise workflows like agentic tasks, RAG, and multilingual document processing. It supports 48 languages, handles text and images, and has a 128,000-token context window. Compared to its predecessor Command A Reasoning, scores jumped from 37 to 85 percent on the agent benchmark τ²-Bench Telecom and from 3 to 25 percent on the coding test Terminal-Bench Hard.
Xiaomi’s new MiMo-V2.5-Pro nearly matches Anthropic’s Claude Opus 4.6 on coding benchmarks while burning 40 to 60 percent fewer tokens, according to the company. The release pushes Xiaomi deeper into the race among Chinese open-weight providers like Deepseek, where the fight is shifting from raw benchmark scores to how cheaply and how long a model can run autonomously on a single task.
US start-up Arcee AI spent roughly half its total venture capital to train Trinity-Large-Thinking, an open reasoning model with 400 billion parameters designed to take on Claude Opus in agent tasks.
Meta is planning to release versions of its new AI models as open source, according to Axios. These would be the first models developed under the leadership of Alexandr Wang, who joined Meta in 2025 as part of a nearly $15 billion deal with Scale AI.
Unlike its approach with the Llama models, though, Meta plans to keep some components proprietary and review safety risks before releasing anything. The largest models won't be made publicly available either.
According to the report, Wang sees Meta as a counterweight to Anthropic and OpenAI, which focus more heavily on government and enterprise customers. Meta's strategy instead centers on consumer reach through WhatsApp, Facebook, and Instagram. Axios's sources say Meta already knows the new models won't match the competition in every area.
