Unsloth
Open-source library that fine-tunes language models several times faster on one GPU.
why this verdict
Keep it — Anthropic has not replaced it
Anthropic's version overlaps, but does not finish the dev tools job, so this one is still worth keeping open.
- A named launch, not a vibe named, unlinked
- Frontier models being good enough that far fewer teams bother fine-tuning anything. Anthropic, September 25, 2025 — no announcement link recorded yet.
- How much of the job it covers not the job editorial call
- Parts of it. The job still needs the tool to get finished.
- Is there a free way to do it? yes
- 2 of 3 listed replacements have a usable free tier: Axolotl and Hugging Face.
- What the call is worth nothing to cancel
- No paid entry tier tracked, so there is no subscription to cancel.
- Threatened by
- Anthropic
- Since
- September 25, 2025
- List price
- free
- Per year
- —
The backstory
Unsloth rewrote the hot paths of fine-tuning in custom kernels, cutting memory use enough that a capable model can be tuned on a single consumer GPU. The macro trend is against fine-tuning, since prompting and retrieval solve most problems now. What keeps Unsloth growing is the opposite end: small open models tuned for one narrow task run far cheaper than a frontier API at volume, and it became the default toolchain the moment anyone decides to try that.
Escape hatches
Config-driven fine-tuning with broad model support
axolotl.ai open_in_newPEFT and TRL libraries with the largest ecosystem
huggingface.co open_in_newManaged fine-tuning if you would rather not run it
together.ai open_in_new2 of 3 replacements have a usable free tier.
