• 3 % discount on prepayment
  • Insured shipping from Switzerland
  • Sustainable: refurbished instead of new
  • Warranty on all cards
← Back to overview

command-a-plus-05-2026-bf16

Cohere
GermanApache 2.0

Open language model for chat and text. 2 sizes from 30.5B to 218.8B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

  • Mixture-of-experts: 218.8B parameters in total, of which 25.5B active per token. Memory requirement depends on all parameters, speed on the active ones.

Settings

Quality
4 bit (Q4_K_M) is the most common choice: good quality with low memory requirements.
Context length: 8k tokens
4k128k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
2× H100 80 GB

The model needs about 133.5 GB of memory. Hopper, FP8 support.

Weights
131.3 GB
KV cache
0.7 GB
Speed
~60.6 Tok./s

More suitable configurations

CardQuantityTok./sStatus
Tesla P40 24 GB8×8.62-3 weeks delivery time
Tesla V100 32 GB8×20.2only 6 in stock
AMD Instinct MI50 32 GB8×15.5only 2 in stock
AMD Instinct MI60 32 GB8×14.32-3 weeks delivery time
AMD Instinct MI100 32 GB8×17.32-3 weeks delivery time
A30 24 GB8×19.62-3 weeks delivery time
RTX A5000 24 GB8×20.42-3 weeks delivery time
A40 48 GB4×24.42-3 weeks delivery time
RTX A6000 48 GB4×27.32-3 weeks delivery time
Quadro RTX 8000 48 GB4×20.22-3 weeks delivery time
L40 48 GB4×30.02-3 weeks delivery time
A100 40 GB4×30.22-3 weeks delivery time