Deploy Open Models on GPUs

Choose a GPU setup and a capability to find the best open models that fit, and deploy one as an API.

Reasoning & Math · 4× NVIDIA RTX 5090

Ranked by benchmark scores for Competition math, logical and scientific reasoning

#	Model	Params	Precision	Max context	Score (0–10)
1	Qwen3.6-27B 2026.04	27.8B	FP16	256K	9.29
2	gpt-oss-120b 2025.08	116.8B	NVFP4	128K	8.90
3	Qwen3.8-27B 2026.08	27.8B	FP16	256K	8.85
4	Qwen3.6-35B-A3B 2026.04	35B	FP16	256K	8.77
5	Baichuan-M2-32B 2025.12	32.8B	FP16	128K	8.34
6	Mistral-Medium-3.5-128B 2026.07	127.7B	NVFP4	186K	8.20
7	Qwen3-Coder-Next 2026.03	80B	FP8	256K	8.20
8	Qwen3.5-122B-A10B 2026.02	122B	NVFP4	256K	7.85
9	Qwen3.5-35B-A3B 2026.02	35B	FP16	256K	7.85
10	Qwen3.5-27B 2026.02	27B	FP16	256K	7.85
11	gpt-oss-safeguard-120b 2025.10	120.4B	NVFP4	128K	7.60
12	Leanstral 1.5 2026.06	119B	NVFP4	256K	7.50
13	Mistral Medium 3.5 2026.04	128B	NVFP4	185K	7.50
14	Ling 2.6 Flash 2026.04	104B	FP8	150K	7.50
15	Mistral Small 4 2026.03	119B	NVFP4	256K	7.50
16	Nemotron 3 Super 2026.03	120B	NVFP4	976K	7.50
17	GLM-4.7-Flash 2026.01	31B	FP16	198K	7.23
18	Leanstral 1.0 2026.03	119B	NVFP4	256K	7.20
19	Qwen-AgentWorld-35B-A3B 2026.06	34.7B	FP16	190K	7.10
20	TaskMem 2026.06	31.1B	FP16	219K	7.10

Deployment verified on HyperAI

Scores aggregate public benchmarks for the selected capability.
Fits usable VRAM with weights plus KV cache at 64K+ context, at the best precision.