llmrun
deepcogitoDeepseekV3ForCausalLM

Cogito 671B V2.1 — Hardware Requirements & GPU Compatibility

ChatFunctions

Cogito v2.1 671B is Deep Cogito's instruction-tuned mixture-of-experts model featuring 671 billion total and 37 billion active parameters, fine-tuned from DeepSeek-V3-Base. This hybrid reasoning approach lets the system answer directly or reflect before responding, and it was trained using Iterated Distillation and Amplification. The card notes optimization for coding, STEM tasks, instruction following, and tool calling across over 30 languages. The BF16 checkpoint occupies about 1.3 TB; according to the card, deployment requires at least 8 B200 GPUs or 16 H200s, with an FP8 variant fitting on 8 H200s. Consequently, this is a datacenter-class model rather than one for local use. The card lists a context length of 128k tokens, whereas the configuration specifies 163,840. Released under the MIT license, it allows unrestricted commercial and research applications. Published in October 2025, this model serves as the 671B member of the Cogito v2.1 family.

671B params · MoE37B active164K contextMIT709 downloads33 likes582 quant downloads

Cogito 671B V2.1 runs on a 512 GB+ Mac or a multi-GPU server

Best local pick → Mac Studio M5 Ultra (512 GB) · Q4_K_M · 404.1 GB · ~26 tok/s

Macs from192 GB IQ2_XXS · ~33 tok/s

MIT

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Cogito 671B V2.1

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

vLLM on a multi-GPU server
tensor + pipeline parallel · needs vLLM on every node
$vllm serve deepcogito/cogito-671b-v2.1 --max-model-len 16384 --tensor-parallel-size 8

Size the nodes from the table above

Download: Q4_K_M/cogito-671b-v2.1-Q4_K_M-00001-of-00009.gguf405.39 GB + 8 partsunsloth

install, other ways

Ways to runOther ways

each opens step-by-step

Run in cloud

Rent 6× H100 SXM · Q4_K_M

Estimated hourly price
~$16.14/hr
est., whole server
GPUs
6 GPUs
480 GB of VRAM

How Much VRAM Does Cogito 671B V2.1 Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KRuns on
289.1 GB287.8 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsRuns on Instinct MI350X 288 GB
297.4 GB296.2 GB3-bit small — 3-bit small quantizationRuns on 5× H100 SXM 80 GB
331.0 GB329.7 GB3-bit medium — 3-bit medium quantizationRuns on 5× H100 SXM 80 GB
339.4 GB338.1 GB4-bit · legacy — 4-bit legacy quantizationRuns on 5× H100 SXM 80 GB
406.5 GB405.2 GB4-bit medium — 4-bit medium quantization — most popular sweet spotRuns on 6× H100 SXM 80 GB
482.0 GB480.7 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffRuns on 7× H100 SXM 80 GB
557.5 GB556.2 GB6-bit, very good — 6-bit quantization, very good qualityRuns on 8× H100 SXM 80 GB
674.9 GB673.6 GB8-bit, near-lossless — 8-bit quantization, near-losslessRuns on 6× H200 SXM 141 GB

Selected Q4_K_M

4-bit medium quantization

404.1GBat 16K context

Weights
402.6 GB
KV cache
1.2 GB
Overhead
0.3 GB

Server-class: ~404.1 GB at 16K context · about 6× H100 SXM 80 GB

Which Hardware Can Run Cogito 671B V2.1?

Q4_K_M · 406.5 GB

Cogito 671B V2.1 (Q4_K_M) requires 406.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 529+ GB is recommended. Using the full 164K context window can add up to 7.9 GB, bringing total usage to 414.4 GB. No consumer GPU has enough memory. 2 devices with unified memory can run Cogito 671B V2.1, including Mac Studio M5 Ultra (512 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 128 GB: too small for this model (needs ≥ 186 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is Cogito 671B V2.1?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

Here are all 2 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.

  • SciCode

    Epoch AI · Nov 2025Lab-reported

    41.0%

    #35/74 open

    Lab-reported
  • CritPt

    Epoch AI · Nov 2025Lab-reported

    0.0%

    #54/85 open

    Lab-reported

Alternatives to Cogito 671B V2.1

Similar models

3
ModelScoreVRAMWhy it is similar
DeepSeek V3.2DeepSeek110.4412.7 GBsame size
DeepSeek R1 0528DeepSeek103.7412.2 GBsame size
Mistral Large 3 675B Instruct 2512Mistral AI87.3454.4 GBsame size

Specifications

Publisher
deepcogito
Parameters
671B
Active parameters
37B per token
Architecture
DeepseekV3ForCausalLM
Context Length
163,840 tokens
Vocabulary Size
128,815
Release Date
2025-10-24
License
MIT

Frequently Asked Questions

How much VRAM does Cogito 671B V2.1 need?

Cogito 671B V2.1 requires 406.5 GB of VRAM at Q4_K_M, or 1345.9 GB at BF16. Full 164K context adds up to 7.9 GB (414.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 671B × 4.8 bits ÷ 8 = 402.6 GB

KV Cache + Overhead ≈ 3.9 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 11.8 GB (at full 164K context)

VRAM usage by quantization

406.5 GB
414.4 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Cogito 671B V2.1?

No — Cogito 671B V2.1 requires at least 188.4 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Cogito 671B V2.1?

For Cogito 671B V2.1, Q4_K_M (406.5 GB) offers the best balance of quality and VRAM usage. Q5_K_S (465.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 188.4 GB.

VRAM requirement by quantization

IQ2_XXS
188.4 GB
Q3_K_S
297.4 GB
Q4_1
381.3 GB
Q4_K_M ★
406.5 GB
Q5_K_S
465.2 GB
BF16
1345.9 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Cogito 671B V2.1 on a Mac?

Yes — Mac Pro M2 Ultra (192 GB) and 5 other Macs can run Cogito 671B V2.1, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 406.5 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio M5 Ultra (512 GB).

Can I run Cogito 671B V2.1 locally?

Yes — Cogito 671B V2.1 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

What's the download size of Cogito 671B V2.1?

At Q4_K_M, the download is about 402.60 GB. The full-precision BF16 version is 1342.00 GB. The smallest option (IQ2_XXS) is 184.53 GB.

Which GPUs can run Cogito 671B V2.1?

No single consumer GPU has enough VRAM to run Cogito 671B V2.1 at Q4_K_M (406.5 GB). Multi-GPU or professional hardware is required.

Which devices can run Cogito 671B V2.1?

2 devices with unified memory can run Cogito 671B V2.1 at Q4_K_M (406.5 GB), including Mac Studio (M3 Ultra, 512GB), Mac Studio M5 Ultra (512 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.