K2 Think V2

Institute of Foundation Models · ifm/k2-think-v2

Provider	Institute of Foundation Models
Family	k2-think
Type	llm-reasoning
Status	active
Released	2026-01-08
Updated	2026-08-05
Parameters	72.6B
Open weights	yes

Lineage

Relationship	Model
Is a finetune of	IFM/K2-V2-Instruct

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.

Device	Memory	Bandwidth	Best quality	Weights	tok/s	Fastest
NVIDIA GB200 Grace Blackwell Superchip	372.0 GB	16000.0 GB/s	bf16	145.1 GB	~77.2	~308.8 q4
AMD Instinct MI355X	288.0 GB	8000.0 GB/s	bf16	145.1 GB	~38.6	~154.4 q4
NVIDIA B200	180.0 GB	7700.0 GB/s	fp8	72.55 GB	~74.3	~148.6 q4
Google TPU7x (Ironwood)	192.0 GB	7380.0 GB/s	fp8	72.55 GB	~71.2	~142.4 q4
AMD Instinct MI325X	256.0 GB	6000.0 GB/s	bf16	145.1 GB	~28.9	~115.8 q4
AMD Instinct MI300X	192.0 GB	5300.0 GB/s	fp8	72.55 GB	~51.1	~102.3 q4
NVIDIA H200 SXM	141.0 GB	4800.0 GB/s	fp8	72.55 GB	~46.3	~92.6 q4
NVIDIA H100 NVL	94.0 GB	3900.0 GB/s	q6	54.41 GB	~50.2	~75.3 q4
NVIDIA H100 SXM	80.0 GB	3350.0 GB/s	q6	54.41 GB	~43.1	~64.6 q4
AMD Instinct MI250X	128.0 GB	3200.0 GB/s	fp8	72.55 GB	~30.9	~61.8 q4
Google TPU v5p	95.0 GB	2765.0 GB/s	q6	54.41 GB	~35.6	~53.4 q4
NVIDIA A100 80GB SXM	80.0 GB	2039.0 GB/s	q6	54.41 GB	~26.2	~39.3 q4
NVIDIA H100 PCIe	80.0 GB	2000.0 GB/s	q6	54.41 GB	~25.7	~38.6 q4
AMD Instinct MI210	64.0 GB	1600.0 GB/s	q5	45.34 GB	~24.7	~30.9 q4
Apple M3 Ultra	512.0 GB	819.0 GB/s	bf16	145.1 GB	~4.0	~15.8 q4
Apple M2 Ultra	192.0 GB	800.0 GB/s	fp8	72.55 GB	~7.7	~15.4 q4
Apple M4 Max	128.0 GB	546.0 GB/s	fp8	72.55 GB	~5.3	~10.5 q4
Apple M1 Max	64.0 GB	400.0 GB/s	q5	45.34 GB	~6.2	~7.7 q4
Apple M2 Max	96.0 GB	400.0 GB/s	q6	54.41 GB	~5.1	~7.7 q4
Apple M3 Max	128.0 GB	400.0 GB/s	fp8	72.55 GB	~3.9	~7.7 q4
Apple M4 Pro	64.0 GB	273.0 GB/s	q5	45.34 GB	~4.2	~5.3 q4

Where it runs

Taken from the card's availability section.

Platform	Id on that platform
Hugging Face	IFM/K2-Think-V2

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	K2 Think V2	57.0%	2026-09-10 evaluated	independent evaluator	source
critpt	K2 Think V2	0.0%	2026-09-10 evaluated	independent evaluator	source
gdpval_aa	K2 Think V2	0.0%	2026-09-10 evaluated	independent evaluator	source
gpqa_diamond	K2 Think V2	71.31%	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