Unsloth

license

Apache-2.0(OSI, core unsloth pkg)+AGPL-3.0(OSI, optional Unsloth Studio UI)

source

public(unslothai/unsloth)

core-gated

gated(unsloth.ai/pricing sells Unsloth Pro and Enterprise on capabilities the free open-source tier does not have: the free column lists MultiGPU as coming soon while Pro lists Enhanced MultiGPU support and up to 8 GPUs, and Enterprise adds multi-node, full training and 5x faster inference

Open core. Everything published is OSI-licensed, Apache-2.0 for the core package and AGPL-3.0 for the optional Unsloth Studio UI, and the source is public. But unsloth.ai/pricing sells Pro and Enterprise on capabilities the free tier does not have, and those are not a hosted service sold beside the open package, they are builds of the same product. The free column lists MultiGPU as "coming soon" while Pro lists "Enhanced MultiGPU support" and "Up to 8 GPUS support"; Enterprise adds multi-node, "Supports full training" and "5x faster inference"; and Pro is advertised as "20% less memory than OSS", an explicit comparison against the published package. Functionality withheld from the published source is what counts as gating on this scale, which is what keeps an otherwise fully open project one band below open source. One caveat: the pricing page still names Llama 1/2/3 and Mistral/Gemma and looks stale, so the exact feature line may have moved even if the tiering has not.

• https://github.com/unslothai/unsloth recorded 2026-08-11

  Live public repository (HTTP 200); GitHub's own page describes it as "The local UI to run and train text and diffusion models, including Kimi K3, Gemma 4, Qwen3.6, DeepSeek-V4, FLUX and more." and lists both an Apache-2.0 license and an AGPL-3.0 license for the repo.

• https://raw.githubusercontent.com/unslothai/unsloth/main/LICENSE recorded 2026-08-11

  LICENSE body states: "Files under unsloth/*, tests/*, scripts/* are Apache 2.0 licensed. Files under studio/*, unsloth_cli/* which is optional to install are AGPLv3 licensed."

• https://unsloth.ai/pricing recorded 2026-08-11

  Three tiers. Free (Open-Source) lists "Open-source", "Supports Mistral, Gemma", "Supports LLama 1, 2, 3", "MultiGPU - coming soon", "Supports 4 bit, 16 bit LoRA". Unsloth Pro lists "2.5x number of GPUs faster than FA2", "20% less memory than OSS", "Enhanced MultiGPU support", "Up to 8 GPUS support". Unsloth Enterprise lists "32x number of GPUs faster than FA2", "up to +30% accuracy", "5x faster inference", "Supports full training" and multi-node support. Both paid tiers are contact-for-pricing.

~2.19M PyPI downloads over the trailing month, down slightly from an earlier reading of ~2.43M, with ~69.7K GitHub stars corroborating. Download volume puts reach in the 1M-10M band.

• https://pypistats.org/packages/unsloth recorded 2026-08-09

  Downloads last month: 2,193,618

• https://github.com/unslothai/unsloth recorded 2026-08-09

  Star button carries aria-label "69730 users starred" this repository, corroborating continued usage volume.

Best-in-class efficiency on a single or consumer GPU, driven by custom Triton kernels that deliver roughly 2x faster training with up to 70% less VRAM, and MoE training around 12x faster with 35% less VRAM. Method coverage is broad too: full fine-tuning, LoRA and QLoRA, RL including GRPO and FP8 RL, pretraining and embedding fine-tuning, across 500+ supported models and several quantization formats. The edge is memory and speed on one node rather than multi-node frontier-scale parallelism, so it sits one band below Megatron-LM.

• https://raw.githubusercontent.com/unslothai/unsloth/main/README.md recorded 2026-08-09

  States "Train and RL 500+ models up to 2x faster with 70% less VRAM; MoE up to 12x faster" and "Supports LoRA/QLoRA, full fine-tuning, RL, pretraining, 4-bit, 16-bit and FP8", plus GRPO/FP8 reinforcement learning and multi-GPU training support, corroborating the recorded feature matrix.