Apple sues OpenAI, and the io deal is the crime scene
Apple's trade secret suit names OpenAI's hardware chief and claims candidates were told to bring actual Apple parts to job interviews. The AI phone might be built on Cupertino's homework.
⚡ Stream
News
Signal over noise: what actually matters in AI and 3D printing this week.
40 news items
China's open-weight Kimi K3 reopens the compute-advantage fight
Moonshot AI's Kimi K3 shipped with open weights and enough capability to restart the DeepSeek argument: if a model you can download keeps pace with the frontier, what were the export controls buying?
llama.cpp b10069: Adreno OpenCL broadcast support
llama.cpp build b10069 adds OpenCL broadcast support for Adreno MUL_MAT operations and fixes view offset handling for Adreno Q8_0 MUL_MAT in llama-server multi-stream mode.
NInfer: 542 tok/s on Qwen3.6-35B-A3B, single request, one RTX 5090
A solo dev open-sourced NInfer, a from-scratch C++/CUDA engine that sustains 542 tok/s on Qwen3.6-35B-A3B over a full 65K-token decode, single request, on one RTX 5090.
Kimi K3 tops frontend code rankings but trails badly on hard math
Moonshot's Kimi K3 sits at #1 on the Code Arena Frontend leaderboard, but scores only ~39% on FrontierMath Tier 4 versus ~90% for top OpenAI and Anthropic models.
Qwen3.6 27B eats your context before the session warms up
Running Qwen3.6 27B with Opencode? The model's appetite for context tokens can gut your usable window fast. Here's what's happening and how to push back.
Fable 5 stays in Max and Team Premium at half limits; Pro gets a $100 credit
From July 20, Claude Fable 5 is included in Max and Team Premium plans at 50% of limits. Pro and Team Standard stay on usage credits at $10/$50 per million tokens, softened by a one-time $100 credit.
Moonshot's Kimi K3 takes #1 on the Frontend Code Arena, beating Fable 5
Kimi K3 hit #1 on the Frontend Code Arena on July 16, beating Claude Fable 5. David Sacks calls it concerning, five weeks after backing the order that benched that same model.
Kimi K3 is a 2.8T open-weight MoE, but the weights are not out yet
Moonshot AI announced Kimi K3, a 2.8-trillion-parameter MoE model, on July 16. Open weights are promised by July 27. Until then it is API-only, and running it yourself will need serious cluster hardware.
llama.cpp b10067 fixes DeepSeek-V4 quantization crash in llama-quantize
llama.cpp build b10067 patches a bug where the DeepSeek-V4 routing table tensor was being incorrectly quantized, causing a type conversion failure that blocked anyone using llama-quantize on the model.
Running three Qwen3.5 122B sessions on one Mac Studio with an on-disk KV cache
A Reddit user reportedly ran three concurrent Qwen3.5 122B sessions on a single Mac Studio, serving over 93% of prompt tokens from an on-disk KV cache instead of recomputing on GPU.
Google quietly updated Gemma 4 to fix tool-calling bugs and add Flash Attention 4 on Hopper GPUs
Google shipped a silent update to Gemma 4 fixing tool-calling bugs, truncated responses, and enabling Flash Attention 4 on Hopper GPUs for a reported 25–70% prefill speedup.
llama.cpp CPU offload reportedly tripled DeepSeek V4 Flash speed on a 4060 Ti
A Reddit report claims the 98GB DeepSeek-V4-Flash quant jumped from 2 to 7 t/s on a single 4060 Ti plus CPU, with no hardware changes between two llama.cpp builds.
Prusa INDX Founder's Edition ships: 8-tool toolchanger conversion kit for CORE One starts at $749
One thousand Bondtech Founder's Edition INDX kits are now shipping for the Prusa CORE One. The standard Prusa Edition follows at end of July, with the full first batch out by end of August.
llama.cpp b9993 adds Hunyuan 3 support with MTP speculative decoding
Build b9993 merges PR #25395, adding native hy_v3 architecture support so Tencent's Hunyuan 3 MoE runs locally in llama.cpp, with MTP speculative decoding included.
Spiritbuun's VBR KV cache squeezes more context into 12GB
A llama.cpp fork from spiritbuun adds Variable Bit Rate KV cache quantization, dynamically degrading cache precision only when VRAM pressure demands it.
vLLM v0.25.0 makes Model Runner V2 the default and drops PagedAttention
vLLM v0.25.0 ships MRv2 as the default execution path for all dense models and permanently removes the legacy PagedAttention implementation.
Colibrì runs GLM-5.2 (744B MoE) on 25GB of RAM by streaming experts from NVMe
Colibrì is a pure-C, zero-dependency inference engine that runs GLM-5.2, a 744B MoE model, on a consumer machine with 25GB of RAM, by keeping dense weights in RAM and streaming routed experts from disk.
Joshua Bird built a ceramic 3D printer using an air-compressor clay extruder
Joshua Bird's custom clay extruder uses an air compressor and stainless steel auger to print pottery. The process works, but shrinkage, abrasion, and a two-fire ceramic workflow make it genuinely hard.
llama.cpp b9963 fixes DeepSeek-OCR v1 multi-tile tiling
Build b9963 lands proper multi-tile support for DeepSeek-OCR v1 with dynamic resolution, and unifies the image preprocessor for both v1 and v2. Multi-column documents now handled correctly.
AMD's $3,999 Ryzen AI Halo undercuts DGX Spark, if you can drive to a Micro Center
AMD's 128GB unified-memory Ryzen AI Halo developer box launched at $3,999, about $700 under NVIDIA's Linux-only DGX Spark, sold in the US exclusively through Micro Center in-store pickup.
Anthropic resets everyone's usage limits and stretches the Fable 5 window to July 12
Claude's 5-hour and weekly limits got a clean-slate reset for all paid tiers this week, and included Fable 5 access was extended from July 7 to July 12 after subscriber backlash.
OpenAI follows GPT-5.6 with ChatGPT Work, an agent that runs your apps
A day after GPT-5.6 went GA, OpenAI shipped ChatGPT Work: an in-ChatGPT agent that works across Slack, Teams, Drive, and email for hours at a stretch, metered like Codex.
GPT-5.6 spent its first twelve days behind a government gate
Before the public launch, GPT-5.6 sat in a preview limited to about 20 government-vetted partners under the new US frontier-AI oversight process. The White House disputes calling the release an approval.