Open Source Factory

MODELS & COSTS Qwen-Image-2.1: A Cheap GPU Estimate Has License Conditions Open Source Factory · Sep 20, 2026 · 10 min read Rented GPU time can make a 10,000-image estimate look inexpensive. That does not buy commercial permission, matching image quality or a finished workflow. Qwen-Image-2.1's research license and the hardware assumptions belong beside the calculation from the beginning. In this article Release capabilities and integrations The components behind the 7B label Memory and dtype requirements A rental-cost estimate for 10,000 images The research-license restriction Quality comparisons and reported failures The release uses the Qwen Research License Agreement, limited to non-commercial research and evaluation in the cited terms. Earlier open Qwen image releases used Apache 2.0. Downloadable weights and familiar tool support do not remove that licensing difference. The model card the day after launch: 888 likes, and the licence tag reads "other" rathe...

MODELS & COSTS Bonsai 2 27B: A Smaller File With Task-Level Trade-offs Open Source Factory · Sep 18, 2026 · 8 min read A roughly 5.9GB model file is easier to place than 54GB of raw half-precision weights. The saving does not preserve every task equally. Bonsai 2's weaker coding rows and its required runtime fork are worth checking before turning the compact download into a local deployment plan. In this article Ternary weights and storage overhead Average retention and the weaker tasks The supported runtime fork Choosing for a particular local task Maintenance, October 17, 2026: a broken reference link has been replaced. Pricing and availability in this article remain the dated snapshot stated in the text; a current documentation link does not turn historical figures into a current offer. I have history with this model's family. In August I wrote about the base model, Qwen3.8-27B , and closed that post promising to revisit it "once independent numbers land and GGUF...

MODELS & COSTS Qwen Max-0902: Pin the Snapshot Before Comparing Open Source Factory · Sep 02, 2026 · 3 min read A familiar API model name can hide a changed checkpoint. Qwen's 0902 update reports coding and office-work gains without a new family name. Pinning the snapshot in a task log makes it possible to ask whether the improvement survives your prompts and acceptance tests. In this article Coding and office-work gains Reading Qwen’s comparison table Price and context conditions API snapshot versus open checkpoint Same model name, new snapshot: "NOW More Capable -0902." (image: Qwen / @Alibaba_Qwen) Coding and office-work gains The August launch was already strong — I flagged TerminalBench and DeepSWE as the highlights, and the open-weights promise landed on schedule (the 2.4T-A95B checkpoint dropped August 12). What 0902 does is fix the embarrassing rows. The agentic terminal coding score that sat at 11.3 in August? 29.0 now. Black-box software replication w...

HARDWARE & DESIGN 35B MoE Memory Planning: What 3B Active Does Not Tell You Open Source Factory · Aug 31, 2026 · 4 min read Updated October 3, 2026 · Author calculations and a planning guide · No hardware benchmark Correction to the earlier version: this URL previously discussed a projected Qwen3.8-35B-A3B release and estimated Mac mini speeds. I could not establish those model-specific speed, context-memory and purchase claims from an official released checkpoint and reproducible measurements. They have been withdrawn. This article now explains the calculation you can check yourself; it does not announce that model or recommend a Mac configuration on the basis of a forecast. Three billion active parameters do not turn a 35B MoE into a 3B download. The inactive experts still need storage, and the application needs memory beyond the weights. Work through that budget before using an active-parameter label to select a 24GB computer. In this article Calculate the raw weights befor...

MODELS & COSTS Qwen Flash-Next: Offload the Table, Budget the Whole Model Open Source Factory · Aug 26, 2026 · 4 min read Flash-Next separates the main MoE from a large n-gram table intended for a host-memory path. That changes which weights are accessed for each token, not whether the full model needs storage. The local question is whether the runtime's offload behavior saves more than the transfers cost. In this article Main MoE and n-gram table Offloading and the full memory budget Qwen's reported benchmark results The separate training-cost claim Qwen Community License Linked source documents That separation changes the memory-planning question. The complete weights still need storage, but they need not all be accessed in the same way for each token. Whether offloading helps a particular computer depends on the runtime and measured transfer overhead. Main MoE and n-gram table Item Qwen3.8-Flash-Next Type Multimodal MoE (image / video / text) Total / active para...