mistralai · text

Mistral-Large-3-675B-Instruct-2512

mistralai/Mistral-Large-3-675B-Instruct-2512

Mistral-Large-3-675B-Instruct-2512 at Q4_K_M is exactly 406,989,520,384 bytes (379.04 GiB / 406.99 GB) — an effective 4.835 bits per weight, not the nominal 4.

From the file· summed from 9 file(s)
Parameters
673B
Architecture
deepseek2
Context
—
native (config.json)
License
apache-2.0

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ1_S4 shards132.50 GiB142,267,169,7601.690—bartowski
IQ1_M4 shards138.29 GiB148,492,663,7761.764—bartowski
IQ2_XXS5 shards153.81 GiB165,151,554,6561.962—bartowski
UD-TQ1_0157.03 GiB168,604,875,2002.003—unsloth
IQ2_XS5 shards176.52 GiB189,541,563,5202.252—bartowski
IQ2_S5 shards177.66 GiB190,766,333,0562.266—bartowski
UD-IQ1_S4 shards177.70 GiB190,801,132,3522.267—unsloth
UD-IQ1_M5 shards192.28 GiB206,460,042,2402.453—unsloth
IQ2_M6 shards201.36 GiB216,213,306,5922.568—bartowski
UD-IQ2_XXS5 shards209.37 GiB224,807,238,6562.671—unsloth
UD-IQ2_M5 shards219.21 GiB235,376,884,6722.796—unsloth
Q2_K7 shards223.36 GiB239,826,713,9522.849—bartowski
Q2_K_L7 shards224.21 GiB240,744,217,9522.860—bartowski
Q2_K5 shards230.14 GiB247,110,065,1522.936—unsloth
Q2_K_L6 shards230.34 GiB247,330,266,2082.938—unsloth
IQ3_XXS7 shards251.10 GiB269,619,903,8083.203—bartowski
IQ3_XS8 shards259.96 GiB279,130,750,4323.316—bartowski
UD-IQ3_XXS6 shards260.04 GiB279,215,127,6163.317—unsloth
Q3_K_S6 shards271.81 GiB291,854,199,8723.467—unsloth
Q3_K_S8 shards275.05 GiB295,337,541,0243.509—bartowski
IQ3_M8 shards288.58 GiB309,858,876,8643.681—bartowski
Q3_K_M8 shards288.62 GiB309,902,917,0563.682—bartowski
Q3_K_L9 shards299.59 GiB321,679,081,0243.821—bartowski
Q3_K_M7 shards299.74 GiB321,845,570,7203.823—unsloth
IQ4_XS8 shards335.23 GiB359,954,316,7364.276—unsloth
IQ4_XS10 shards337.09 GiB361,952,919,2004.300—bartowski
IQ4_NL8 shards354.59 GiB380,738,813,2484.523—unsloth
Q4_08 shards355.49 GiB381,700,357,4404.535—unsloth
IQ4_NL10 shards356.18 GiB382,449,958,6244.543—bartowski
Q4_K_S8 shards356.38 GiB382,661,901,6324.546—unsloth
Q4_010 shards361.71 GiB388,388,044,4804.614—bartowski
Q4_K_S11 shards368.91 GiB396,117,098,3364.706—bartowski
Q4_K_M9 shards379.04 GiB406,989,520,3844.835—unsloth
Q4_K_M11 shards383.10 GiB411,354,713,9524.887—bartowski
Q4_19 shards393.36 GiB422,366,526,9125.018—unsloth
Q4_111 shards393.95 GiB422,996,295,4245.025—bartowski
Q5_K_S10 shards432.55 GiB464,446,570,0805.518—unsloth
Q5_K_M10 shards445.14 GiB477,969,661,5365.678—unsloth
Q5_K_M13 shards447.75 GiB480,762,839,1685.711—bartowski
Q6_K12 shards515.32 GiB553,317,879,6166.573—unsloth

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs Qwen3.8-27Bvs DeepSeek-V4-Flashvs Qwen3-30B-A3B-Thinking-2507vs Laguna-XS-2.1vs DeepSeek-V4-Flash-0731All alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 352.78 GiB. The real file is 379.04 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

Architecture unavailable — this repository is gated and no ungated mirror was found. Exact file sizes above are still authoritative; only the KV math needs the config.

Questions people ask

How much VRAM does Mistral-Large-3-675B-Instruct-2512 need?
Q4_K_M is exactly 406,989,520,384 bytes (379.04 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of Mistral-Large-3-675B-Instruct-2512 should I use?
Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.