INTELLECT-3

Prime Intellect · primeintellect/intellect-3

Provider	Prime Intellect
Family	intellect
Type	llm-chat
Status	active
Released	2025-11-26
Updated	2025-11-27
Parameters	106.9B
Open weights	yes

Lineage

Relationship	Model
Is a finetune of	zai-org/GLM-4.5-Air-Base

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices. This is a mixture-of-experts model and no card carries active-parameter counts, so these predictions use total parameters and understate the real speed.

Device	Memory	Bandwidth	Best quality	Weights	tok/s	Fastest
NVIDIA GB200 Grace Blackwell Superchip	372.0 GB	16000.0 GB/s	bf16	213.7 GB	~52.4	~209.6 q4
AMD Instinct MI355X	288.0 GB	8000.0 GB/s	bf16	213.7 GB	~26.2	~104.8 q4
NVIDIA B200	180.0 GB	7700.0 GB/s	fp8	106.85 GB	~50.4	~100.9 q4
Google TPU7x (Ironwood)	192.0 GB	7380.0 GB/s	fp8	106.85 GB	~48.3	~96.7 q4
AMD Instinct MI325X	256.0 GB	6000.0 GB/s	fp8	106.85 GB	~39.3	~78.6 q4
AMD Instinct MI300X	192.0 GB	5300.0 GB/s	fp8	106.85 GB	~34.7	~69.4 q4
NVIDIA H200 SXM	141.0 GB	4800.0 GB/s	q6	80.14 GB	~41.9	~62.9 q4
NVIDIA H100 NVL	94.0 GB	3900.0 GB/s	q5	66.78 GB	~40.9	~51.1 q4
NVIDIA H100 SXM	80.0 GB	3350.0 GB/s	q4	53.43 GB	~43.9	~43.9 q4
AMD Instinct MI250X	128.0 GB	3200.0 GB/s	q6	80.14 GB	~28.0	~41.9 q4
Google TPU v5p	95.0 GB	2765.0 GB/s	q5	66.78 GB	~29.0	~36.2 q4
NVIDIA A100 80GB SXM	80.0 GB	2039.0 GB/s	q4	53.43 GB	~26.7	~26.7 q4
NVIDIA H100 PCIe	80.0 GB	2000.0 GB/s	q4	53.43 GB	~26.2	~26.2 q4
Apple M3 Ultra	512.0 GB	819.0 GB/s	bf16	213.7 GB	~2.7	~10.7 q4
Apple M2 Ultra	192.0 GB	800.0 GB/s	fp8	106.85 GB	~5.2	~10.5 q4
Apple M4 Max	128.0 GB	546.0 GB/s	q6	80.14 GB	~4.8	~7.2 q4
Apple M2 Max	96.0 GB	400.0 GB/s	q5	66.78 GB	~4.2	~5.2 q4
Apple M3 Max	128.0 GB	400.0 GB/s	q6	80.14 GB	~3.5	~5.2 q4

Where it runs

Taken from the card's availability section.

Platform	Id on that platform
Hugging Face	PrimeIntellect/INTELLECT-3

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.

Benchmark	Evaluated as	Score	Evidence date	Source kind
aa_lcr	INTELLECT-3	39.67%	2026-09-10 evaluated	independent evaluator	source
arena_elo_style_control	intellect-3	1356.16 elo	2026-09-10 evaluated	independent evaluator	source
critpt	INTELLECT-3	0.29%	2026-09-10 evaluated	independent evaluator	source
gpqa_diamond	INTELLECT-3	76.06%	2026-09-10 evaluated	independent evaluator	source

Reported benchmark scores

This card reports no benchmark scores yet.

Benchmark	Catalogue standing	Score

Data

This card as JSON · See it in the graph · Edit on GitHub