oMLX remains under the Apache 2.0 license in the same repository, with Jun Kim continuing to lead it while working on the project full time at Hugging Face. He says the effort began as a personal project in February.
Kimi.ai says Kimi K3 is available on Amazon Bedrock for coding, document analysis, and extended agent workflows, with Bedrock access, encryption, and auditing controls and support for explicit prompt caching.
StepFun highlighted Artificial Analysis results putting Step 5 Preview at 44 on the Intelligence Index and $0.71 per task, with more to come on Oct 15. The evaluation describes a 600B/27B MoE flagship with first-party API access and open weights planned for October 15.
Tencent Hunyuan said Hy Image3.5 preview is live, with text-to-image and image-to-image generation up to 2K. Tencent Cloud API pricing is $0.024 per generated image, and it offered two weeks free only on OnSolo and Miora.
Tencent Hunyuan quoted a Zhihu Frontier explainer on shrinking Hy4 preview's 770B-parameter weights from about 1.5TB to 214 GiB with mixed-precision quantization. The quoted write-up says the parameter count is unchanged and reports nearly unchanged MRCR with a small decline in math.
Command Code said Grok 4.7 is now live in Command Code and available on GOAT, Pro, Max, and via API. It also listed 40% for the next 7 days and $35 credits (3.5x usage) in GOAT.
Dax (@thdxr) said Kit finished a blog post covering how OpenCode 2.0 can modify everything about itself in real time, even during running sessions.
TypeSafe quoted MotherDuck’s claim that prompt_jev(), a SQL function powered by TypeSafe’s Jev model, made text classification about 50x faster at about 1% of the cost. MotherDuck reported 100k rows in 40 seconds for $0.50 versus 32 minutes and $37 for an LLM.
Alibaba Cloud and ZAKA are coming together for a two-day Qoder AI Hackathon at AI Everything Abu Dhabi 2026 on 6–7 October at ADNEC Centre Abu Dhabi, where participants can build working AI solutions with Qoder.
Command Code said MiMo V2.6 Pro and Flash are live in Command Code and available on all plans. On the $10 GOAT plan, it listed MiMo-V2.6 Flash at $67 (6.7x usage) for 72hrs and MiMo-V2.6 Pro at $20 (2x usage).
Vercel said to build with Grok 4.7 and Vercel at a lower cost, with 40% off on AI Gateway until September 27th. The post also said to try it with @v0, @eve, and fx.
The one-time method splits a question into clues, runs complementary searches, and reranks the pooled results before the agent loop starts. On BrowseComp-Plus, with the same retriever and agent loop, GPT-5.5 rises from 83.1% to 90.5%.
Cline said flagship model MiMo-v2.6-Pro is officially out and already available in ClinePass. It also called the model the best-performing open-weights model on the artificial intelligence index.
Mark Zuckerberg said Muse is teaming up with Shopify to make shopping and checkout easier in Muse. He also said more partnerships like this are coming soon.
v0 said Grok 4.7 is now available in v0, with 40% off until September 27th. The post quoted SpaceXAI’s announcement that Grok 4.7 is here.
Google said last week it introduced two new audio models from Google DeepMind, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, and that team members are building with them across creative applications, real-time support, and robotics.
OpenCode said MiMo V2.6 Flash is free for the next week. The same post said both Flash and Pro are also available in Go.
vLLM says it has integrated PyNvVideoCodec to offload video decoding from CPU to GPU NVDEC, reporting 2x+ throughput at 8×H100. The project says the change ships with CUDA vLLM releases.
Grok posted “Try Grok 4.7 in Box,” quoting Box’s preview of a Box Agent that uses Grok 4.7 to reconcile a $2 million insurance claim against policy records and produce a cited review for an adjuster.
SambaNova said SN50 is built for premium inference by pairing RDU decode with existing GPUs for prefill. It cited ROI in 6 months at a 70% billable utilization planning assumption.
