# M3 Published articles for M3. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Asahi Linux Makes Progress On Gaming, More Apple Silicon Upstreaming DevFeed: [Asahi Linux Makes Progress On Gaming, More Apple Silicon Upstreaming]() Original publisher: [Read original article]() Author: Michael Larabel Published: 2026-11-07T10:21:03Z Content type: news Language: en Sources: [Phoronix]() Topics: [Linux Kernel](), [Linux](), [Reverse Engineering](), [Phoronix]() Tags: [apple-silicon](), [audio](), [desktop-linux](), [gaming](), [gpu-acceleration](), [linux](), [linux-benchmarking](), [linux-hardware-benchmarks](), [linux-hardware-reviews](), [linux-how-to](), [linux-kernel](), [linux-performance](), [linux-server-benchmarks](), [m3](), [open-source-graphics](), [phoronix](), [phoronix-test-suite](), [thunderbolt](), [ubuntu-benchmarks](), [ubuntu-hardware](), [wine]() ### AI overview Asahi Linux reports continued Apple Silicon support work, including upstream audio changes, initial device trees for M4 Pro, M4 Max, and M5, and Thunderbolt support submitted for review. Downstream Apple M3 support is now considered generally available, but lacks GPU acceleration; developers have rendered a triangle, with substantial work still ahead. Gaming is becoming more feasible on M1 and M2 using EX and Wine, and work is underway to run games outside muvm to handle kernel page-size differences. Performance with newer games remains patchy. ### Source excerpt Given the recent Linux 7.3 upstream kernel release, the Asahi Linux project published a new status report to outline recent code they have upstreamed to the mainline kernel as well as other efforts taking shape in their downstream codebase... ## MiniMax M3.1 Flash vs. MiniMax M3: One is cheaper, but not faster or better DevFeed: [MiniMax M3.1 Flash vs. MiniMax M3: One is cheaper, but not faster or better]() Original publisher: [Read original article]() Author: Jessica Wachtel Published: 2026-10-26T13:30:00Z Content type: comparison Language: en Sources: [The New Stack]() Topics: [Benchmark](), [Cache](), [Test automation]() Tags: [2026](), [agent-tasks](), [agentic](), [agentic-tasks](), [ai-agents](), [ai-models](), [artificial-analysis](), [automationbench](), [benchmark](), [cost](), [large-language-models](), [m3](), [m3-1-flash](), [minimax](), [minimax-m3-1-flash](), [october](), [software-testing](), [vs]() ### AI overview MiniMax M3.1 Flash cost less than M3 in the author's three one-off tests, but did not show a clear overall improvement: it was slower overall, tied M3 on both graded tasks, and neither model reliably answered the long-context task. In a tool-based bug fix, Flash corrected all three bugs but used its 25-turn limit before reporting a wrong test; M3 finished faster but missed a case. In an incident investigation, both scored 10/10, while Flash completed in fewer turns and at lower estimated cost. Flash's first long-context run exhausted the output allowance; with a lower limit it answered only 3 of 33 affected services, while M3 again failed to answer. Costs were estimates at list prices, and results came from one run per task through OpenRouter, so they do not establish general performance or cost differences. ### Source excerpt MiniMax released M3.1 Flash on October 19, 2026, and says it cuts the overall cost of agentic tasks by 70% The post MiniMax M3.1 Flash vs. MiniMax M3: One is cheaper, but not faster or better appeared first on The New Stack. ## Raspberry Pi 5 Case - Pimoroni NVMe Base + Noctua 40x10mm Fan #3DThursday #3DPrinting DevFeed: [Raspberry Pi 5 Case - Pimoroni NVMe Base + Noctua 40x10mm Fan #3DThursday #3DPrinting]() Original publisher: [Read original article]() Author: Pedro Published: 2026-10-23T00:00:42Z Content type: article Language: en Sources: [Adafruit Industries]() Topics: [Raspberry Pi](), [3D]() Tags: [3d](), [3d-printing](), [3dprinting](), [3dthursday](), [adafruit](), [arduino](), [build](), [cad](), [community](), [diy](), [download](), [m3](), [noctua](), [nvme](), [pdf](), [pi](), [pi-5](), [pimoroni](), [raspberry-pi](), [raspberry-pi-5](), [ssd]() ### AI overview The case is designed for a Raspberry Pi 5 with the Pimoroni NVMe Base and a Noctua 40x10 mm 5 V fan. It comes in two versions with the same lid: one uses Ruthex RX-M3x5.7 brass threaded inserts, while the other takes M3 screws directly. Both versions use M3 countersunk screws. A PDF guide is linked on the model page. ### Source excerpt user_258400175 shares: A Pi 5 case for the Pimoroni NVMe Base (M.2 SSD) + a Noctua 40x10mm 5Vdc fan. Two versions share the same lid: one with M3 threaded inserts (Ruthex RX-M3x5.7 brass), one that uses M3 screws directly without inserts. Both use M3 countersunk screws. Includes a PDF guide (Raspberry Pi5 Case.pdf) linked on [...] ## vLLM Delivers Day-0 Support for MiniMax-M3.1-Flash DevFeed: [vLLM Delivers Day-0 Support for MiniMax-M3.1-Flash]() Original publisher: [Read original article]() Author: Inferact and the vLLM Team Published: 2026-10-23T00:00:00Z Content type: release Language: en Sources: [vLLM Blog]() Topics: [vllm](), [Model Optimizer](), [NVFP4](), [quantization](), [rocm](), [NVIDIA Vera Rubin]() Tags: [ai-engineering](), [amd](), [blackwell](), [day-0-support](), [decoding](), [dspark](), [flash](), [gpus](), [hardware](), [kernels](), [llm-inference](), [m3](), [m3-1-flash](), [minimax](), [minimax-m3-1-flash](), [model-serving](), [nvfp4](), [nvidia](), [nvidia-blackwell](), [nvidia-rubin](), [open-source-ai](), [performance](), [rl](), [series](), [support](), [vllm](), [vllm-blog]() ### AI overview vLLM provides day-0 MiniMax-M3.1-Flash support on NVIDIA Blackwell and Rubin and AMD MI350-series GPUs, with NVFP4 KV cache and MoE support, DSpark speculative decoding, and RL training from the NVFP4 checkpoint. The model's sparse attention selects at most 16 blocks of 128 tokens per query, limiting attention to 2,048 keys; the article estimates long-context decode HBM traffic at about 12.5% of dense attention. On four B300s, NVFP4 reduces checkpoint size 44.6% and KV-cache bytes per token 43.75%; at 256K context, the stated deployment fits 76 sequences versus 28 for the prior model's FP8 cache. Hardware-specific optimizations include CUDA 13.4 QMUL4 kernels on Blackwell, distinct MSA kernels on Rubin, and table-based NVFP4 conversion, context parallelism, and fused low-latency decoding on AMD. Reported AgentX results vary by platform and metric: Rubin reaches 724.9 tokens/s per user at the latency-sensitive frontier, while AMD MI355X reaches 133.3M total tokens per dollar at the cost-efficient frontier. Day-0 RL uses veRL, FSDP2, rank-32 LoRA, and token-level importance sampling; a 16-GB300, 200-step run raises AIME-2024 accuracy from 36.3% to 67.5%. Prebuilt NVIDIA and AMD vLLM- ### Source excerpt Day-0 vLLM support for MiniMax-M3.1-Flash on NVIDIA Blackwell, NVIDIA Rubin, and AMD MI350-series GPUs, with NVFP4 MSA kernels, NVFP4 MegaMoE, DSpark speculative decoding, AgentX results, and day-0 RL. ## Arduino Uno R4 WiFi + Minima Case - Several Top-Case Connector Variants #3DThursday #3DPrinting DevFeed: [Arduino Uno R4 WiFi + Minima Case - Several Top-Case Connector Variants #3DThursday #3DPrinting]() Original publisher: [Read original article]() Author: Pedro Published: 2026-10-22T11:00:11Z Content type: article Language: en Sources: [Adafruit Industries]() Topics: [3D](), [Uno R4](), [Arduino]() Tags: [3d](), [3d-printed](), [3d-printing](), [3dprinting](), [3dthursday](), [arduino](), [assembly](), [connectors](), [m3](), [plate](), [recommended](), [requires](), [shares](), [uno](), [uno-r4](), [wifi]() ### AI overview The case fits the Arduino Uno R4 WiFi or Minima and comes with top-case variants for different connector needs. Holes allow it to be mounted to a plate. It can hold together without screws; screw assembly uses M3 threaded inserts and M3x12 screws. Suggested print settings are 0.2 mm layers, 3 walls, and 30% infill, with the excerpt estimating about 1.2 hours and 32 g of PLA. ### Source excerpt tim290923 shares: A case for the Arduino Uno R4 WiFi or Minima. Several top-case variants depending on which connectors you need. Can be mounted and secured to a plate through its holes. Screw assembly requires M3 threaded inserts and M3x12 screws, but holds together without screws. Recommended: 0.2mm layer, 3 walls, 30% infill, ~1.2h, 32g [...] ## MiniMax launches M3.1 Flash model, says it cuts AI agent costs by 70% DevFeed: [MiniMax launches M3.1 Flash model, says it cuts AI agent costs by 70%]() Original publisher: [Read original article]() Author: Duncan Riley Published: 2026-10-19T22:25:50Z Content type: news Language: en Sources: [SiliconANGLE]() Topics: [LLM optimization](), [Artificial Intelligence](), [Large Language Model](), [AI Agent](), [LLM evaluation / benchmarking](), [AI Chat]() Tags: [agent](), [agent-costs](), [agentic-tasks](), [ai](), [ai-agents](), [api](), [bankertoolbench](), [benchmark](), [berkeley](), [cybergym](), [cybersecurity](), [deepseek](), [deepswe](), [hangzhou-deepseek](), [hong-kong](), [inference](), [initial-public-offering](), [large-language-model](), [large-language-models](), [m-plan](), [m3](), [m3-1-flash](), [minimax](), [minimax-code](), [model-weights](), [news](), [openrouter](), [osworld-v2](), [reasoning](), [release](), [space-bunny](), [sparsity](), [the-latest](), [top-story-3](), [top-story-6](), [university-of-california](), [v4-1-flash]() ### AI overview MiniMax released M3.1 Flash, a faster, lower-cost update to its M3 large language model aimed at reducing the cost of running AI agents. The company reports 70% lower agentic-task costs and inference speeds close to three times faster than its previous model, along with benchmark results for computer operation, banking workflows, cybersecurity, and software engineering. The model is available through MiniMax's API, subscription plan, and desktop app; the company says it plans to release the weights later. ### Source excerpt Chinese artificial intelligence developer MiniMax Group Inc. today released M3.1 Flash, a faster and cheaper update to the M3 large language model it launched in June. The release takes aim at the cost of running AI agents, which often need dozens of model calls to finish a single job. MiniMax said Flash represents its "ongoing [...] The post MiniMax launches M3.1 Flash model, says it cuts AI agent costs by 70% appeared first on SiliconANGLE. ## Linux Kernel 7.3 Release is Good News For Gamers Short on VRAM DevFeed: [Linux Kernel 7.3 Release is Good News For Gamers Short on VRAM]() Original publisher: [Read original article]() Author: Sourav Rudra Published: 2026-10-19T02:56:37Z Content type: news Language: en Sources: [It's FOSS]() Topics: [Linux Kernel](), [Kernel]() Tags: [apple](), [kernel](), [linux](), [linux-kernel](), [m3](), [memory](), [news](), [release](), [support](), [vram](), [zen-6]() ### AI overview Linux 7.3 adds an initial kernel change to help games run on GPUs short on VRAM: the kernel can move less important data to system RAM to free video memory. The article says the full patch set is still on its way to Linux, though it already ships in SteamOS; with all patches, Indiana Jones and the Great Circle ran well enough to play on an 8GB card despite requesting 9GB. Other reported changes include KSM lock improvements that cut a stress-test worst case from 705ms to under 2ms, and zsmalloc gains of up to 1.83x on Raspberry Pi 4B and 1.4x on a 20-core Intel system. The release also adds Zen 6 groundwork, initial Linux boot support for Apple M3 Pro, Max, and Ultra, and assorted device support. The article identifies Ubuntu 26.10 as a straightforward upgrade route; installing the mainline kernel on Ubuntu carries a data-loss risk, so it advises backing up first. ### Source excerpt A fix for memory freezes lands alongside Zen 6 groundwork, and early Apple M3 support.