1× RTX PRO 6000
207tok/sMixed
1 request, mixed prompt set, with FR-Spec + NEXTN (per recipe) Source
Intelligence40with thinking
- Custom kernel required
- License unclear
Local AI · Recipe · 1× RTX PRO 6000
K2-Horizon 36B-A4B EXL3 8.00 bpw with ExLlamaV3 on 1× RTX PRO 6000
K2-Horizon 36B-A4B EXL3 8.00 bpw with ExLlamaV3 on 1× RTX PRO 6000: 30 tok/s according to github.com (dataset as of Nov 3, 2026).
by vcruz305
Engine
- Engine
- ExLlamaV3 + TabbyAPI Fork be01a273, TabbyAPI f07131cd
- Quantization
- EXL3 2.00–8.00bpw
- Model family
- K2-Horizon (IFM)
- Context
- 65,536
- Parameters
- 36B / 4B active
- Creator
- vcruz305
- GitHub stars
- 1
- Repo updated
- Sep 23, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Prefill by context
Time to first token
Shorter is better.
What you need
- Hardware
- 1 × NVIDIA RTX PRO 6000 Blackwell
- Weights
- Main weightsvcruz305/K2-Horizon-MoVA-36B-A4B-EXL3 Apache-2.0
- BaseIFM/K2-Horizon-MoVA-36B-A4B Apache-2.0
- Engine
- ExLlamaV3 + TabbyAPI Fork be01a273, TabbyAPI f07131cd
- Context
- 65,536 tokens
Notes
What matters before you rebuild it.
- Custom kernel required
Sources
Related recipes
1× RTX PRO 6000
287tok/sPeak
1 request, synthetic prompt Source
Intelligence9with thinking
1× RTX PRO 6000
281tok/sPeak
1 request, synthetic prompt Source
Intelligenceno independent value
1× RTX PRO 6000
240tok/sPeak
1 request, best value reported by source Source
Intelligence34with thinkingno thinking 20
1× RTX PRO 6000
196tok/sPeak
1 request, synthetic prompt Source
Intelligence12with thinking
1× RTX PRO 6000
184tok/sPeak
1 request, best value reported by source, context 64k Source
Intelligenceno independent value
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.