Open Source Factory

AGENTS & WORKFLOWS MiniMax M3.1 Flash: Price the Finished Task Open Source Factory · Oct 20, 2026 · 6 min read Watch the English motion graphic 48-second motion film. Animation, newly composed music and sound design: Open Source Factory. Sources: MiniMax and Plotly. Conceptual animation, not a model-run recording. A low token rate makes an agent cheap to start. It can still spend the saving on repeated searches, long answers or failed attempts. For MiniMax M3.1 Flash, the useful calculation follows the entire job through to an accepted result. In this article What can you actually use today? What does the rate card buy? Does cheap pricing survive a whole task? How would I test it on my own work? Count the work after the first call MiniMax M3.1 Flash is worth trying for work that can run in the background. Its official API rates are low, and an outside analytics evaluation gives it a credible result. But the bill for a completed job depends on how much work the agent creates alon...

IMAGE, VIDEO & AUDIO MiniMax Music 3: Control, Hardware and License Conditions Open Source Factory · Aug 13, 2026 · 6 min read Section tags and downloadable weights give Music 3 a workflow to inspect beyond the launch samples. The commercial-license conditions and GPU path are just as concrete. A five-minute song limit does not answer whether the released model sounds right for the music you want to make. In this article Released model and controls Global and local architecture Community-license conditions What you need to run it Evidence still missing at launch Is MiniMax Music 3 actually open source? Can I run it on 8GB of VRAM? Can I use the songs commercially? How long can a song be? Is it better than Suno? An 8B global model and 0.6B local model feed a flow-matching decoder. The release uses MiniMax's community license, including a $20 million annual-revenue threshold and a product-labeling requirement. Downloadable weights, commercial permission and unrestricted open-s...

MODELS & COSTS MiniMax H3's 4B Encoder Swap: One 1080 Ti Experiment Open Source Factory · Aug 10, 2026 · 4 min read The encoder swap solved a concrete setup problem on this 11GB card, but the controls are the part worth inspecting. With one prompt and seed, the learned projection produced the intended scene while the zero and identity matrices did not. That is a small result, with a useful boundary. In this article The encoder being replaced The verification: "a red ball on a wood table" Timing and offloading on the 1080 Ti Observed losses and test limits What this experiment supports The encoder being replaced The baseline encoder is Qwen3-VL-32B truncated to 50 layers, with 15.7GB of NVFP4 weights producing a [seq, 5120] conditioning tensor. That exceeds the card's 11GB VRAM before accounting for the video model. It explains why encoder placement was worth testing on this setup; other machines may have a different bottleneck. ClipProj, an MIT-licensed proj...

IMAGE, VIDEO & AUDIO MiniMax H3 Before the Weights: A Memory Estimate's Limits Open Source Factory · Aug 02, 2026 · 4 min read October 3, 2026 update: The “weights coming soon” wording below describes the original announcement. MiniMax subsequently published weights and local-deployment instructions on August 3; use the linked release guide for the current files. Before the H3 files arrived, a capacity calculation could only identify likely constraints. It could not establish a working local pipeline. The later release reports and encoder experiment linked here provide narrower evidence about actual configurations, which should be read separately from the original estimate. In this article Announced service capabilities Launch service pricing The original local-memory estimate Weight-release and license conditions Post-release evidence Linked source documents The initial announcement promised weights within days. This article records the pre-release capacity calculation....