Qwen has released Qwen3.8 2.4T A95B, an open-weight sparse mixture-of-experts model with 2.4 trillion total parameters and 95 billion active parameters per forward pass. The model is the open-weight variant of Qwen3.8 Max, targeting coding, research, complex reasoning, and agentic workflows with a 262K token context window.
DeepSeek has shipped the general availability release of DeepSeek V4 Pro, codenamed 0813, featuring a 1,049,000-token context window. The mixture-of-experts model is priced at $0.43 per million input tokens and $0.87 per million output tokens, and is live now on OpenRouter.
xAI released Grok 4.6, claiming intelligence on par with OpenAI's GPT-5.6 Sol and just one point behind Anthropic's Claude Fable 5 Max on the Artificial Analysis Intelligence Index. The model is priced at $2 per million input tokens and $6 per million output tokens, with a faster variant at double that rate.
Vercel released version 4.0.37 of its @ai-sdk/xai package, adding support for Grok 4.6 model IDs and a new 'xhigh' reasoning effort level. The patch update lets developers using the Vercel AI SDK target xAI's newest model IDs directly.
Vercel released @ai-sdk/xai version 3.0.119, a patch update that adds model IDs for Grok 4.6 and introduces support for an 'xhigh' reasoning effort parameter. The change lets developers using the Vercel AI SDK target xAI's newer model directly through the existing provider interface.
Vercel shipped @ai-sdk/xai version 2.0.86, a patch release that adds support for Grok 4.6 models in its xAI provider integration. The update lets developers using the Vercel AI SDK route requests to Grok 4.6 without custom API handling.
Three independent data sources — Pangram, OpenRouter, and Similarweb — indicate Google's Gemini is losing market share to ChatGPT and Claude. Pangram reports Gemini's share fell from 12% to 1.9% between tracking periods, while Anthropic climbed from 4.3% to 14.9%.
xAI has released Grok 4.6, a text-and-image model featuring a 500K token context window. The model is priced at $2.00 per million input tokens and $6.00 per million output tokens, and is available now through OpenRouter's API.
ByteDance Seed has released Seed 2.1 Turbo, a multimodal model targeting coding and long-horizon agent workflows with a 262K token context window. The model is priced at $0.50 per 1M input tokens and $2.50 per 1M output tokens, and is now listed on OpenRouter.
Alibaba's Qwen team has released Qwen3.8-2.4T-A95B, a 2.4-trillion-parameter mixture-of-experts model activating 95B parameters per token. The company claims it is the first open-weight release to reach Qwen-Max-class performance, with a hosted variant, Qwen3.8-Max, offering 1M-token context and vision input.
Google DeepMind has released SL2T, a massively multilingual sign-language-to-text translation model trained on over 100,000 hours of data across 50+ sign languages. The model powers new sign-to-text dictation features in Gboard and Live Transcribe on Pixel 11, starting with American Sign Language to English.
Liquid AI has released LFM2.5-VL-3B, a 3.1-billion-parameter vision-language model built for on-device deployment. The model claims leading performance in its size class on grounding, screen understanding, and tool use, while running at 228 tokens/s on an Apple M5 Max.
Nvidia is reportedly building Nemotron 4, an open-weight model with at least one trillion parameters — double the size of Nemotron 3 Ultra. The company has tripled its cloud spending on in-house training to $28 billion through 2031, with an earliest possible release this fall.
Microsoft's MAI Code 1.1 Flash beats its predecessor and mini-models from Anthropic and OpenAI on SWE-bench Verified, but DeepSeek-V4-Flash-0731 outperforms it on Terminal Bench 2.1 (82.7% vs 62.9%) while costing roughly a third as much per token.
Mistral has made regional inference generally available, letting customers route requests through EU or US servers for a 10 percent surcharge, while also launching a paid Priority Tier that charges 1.75x standard pricing for faster processing during peak traffic. Both offerings carry significant limitations on what data and features they actually cover.
Anthropic is adding machine-readable watermarks to Claude-generated text and digital signatures to generated files to comply with the EU AI Act's Article 50 transparency mandate. The change applies to models launched after Aug. 2 and rolls out globally, though Anthropic admits detection can fail on heavily edited or short text.
Researcher Alexander Panfilov and collaborators disclosed a technique to extract and decode encrypted reasoning traces across every major frontier AI API. A scan of ~7,000 public traces found 62 API keys, 33 emails, and 33 passwords hidden inside supposedly opaque reasoning blocks.
Vercel released @ai-sdk/xai version 4.0.36, adding support for xAI's grok-imagine-video-1.5 model with native 1080p resolution and a new referenceVoiceIds parameter for reference-to-video audio. The update also fixes a routing bug that misdirected certain video generation requests.
A paper published at stolen-thoughts.com demonstrates that encrypted reasoning traces returned by OpenAI, Anthropic, and Google APIs used the same encryption key across models in a family, allowing attackers to jailbreak weaker sibling models into revealing a stronger model's hidden chain-of-thought in plaintext. All three providers have since patched the vulnerability.
NVIDIA has published NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4, a 30-billion-parameter Mixture-of-Experts model with only 3B active parameters, a hybrid Mamba-2/MoE/Attention architecture, and support for up to 1 million tokens of context. The NVFP4-quantized checkpoint is designed to run on a single DGX Spark (GB10) or H100 GPU.