Thank you for releasing this model. I've been testing several of your Mellum releases, and they've consistently been among the most interesting reasoning-focused models available in the open-source community.
I especially appreciated the earlier abliterated Mellum release, which demonstrated that the model can remain highly capable while being more flexible for research, creative writing, roleplay, agent workflows, and advanced user customization.
Would you consider releasing an official abliterated or less-restrictive version of this Claude-tuned Mellum model as well? It would be very interesting to compare the reasoning capabilities of the original release with a research-oriented variant.
It could also be interesting to see collaboration or community contributions from people such as Huihui, Mradermacher, and Eric Hartford, who have each contributed valuable work to the open-source model ecosystem.
In any case, thank you again for continuing to develop the Mellum series and for sharing these models with the community. I'm looking forward to seeing future releases.
@ridham034 Thank you — that genuinely means a lot, especially on the Mellum line; it's one of the releases I'm most personally fond of, so I'm glad the reasoning focus landed.
On a less-restrictive / abliterated variant: I'm open to it in principle — I've done abliteration work on other models, everything I put out is Apache-2.0, and I like that direction for research, creative and agent use. Honest caveat on timing: I'm fully heads-down on v3 (the Gemma agentic line) plus a new open-source collaboration with a lab right now, so I can't promise a near-term official abliterated Mellum from me.
But here's the good part — because the Claude-tuned Mellum is Apache-2.0, nobody has to wait on me. People like Huihui, Mradermacher, and Eric Hartford can build directly on it (abliteration, imatrix quants, uncensored variants), and I'd genuinely welcome it — happy to boost anything good that comes out of it, and to compare reasoning against the original like you suggested. If any of them, or you, want to take a run at it, please do — and tag me.
Thanks again for such a thoughtful note; it's the kind of thing that keeps me building. 🙏
