readygpu

Devstral Small 2507

23.6B parameters · 128k context · Apache 2.0 · released 2025-07-04

codingagentic

In short

Devstral Small 2507 is an open model from Mistral AI with 23.6B parameters. At Q4_K_M, Devstral Small 2507 needs about 15 GB of memory at 8k context and 19 GB at 32k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in the US that runs Devstral Small 2507 fully in memory at 32k context is the AMD Radeon RX 7900 XTX, $1,533 at Amazon, at about 24 tok/s (estimated).

Updated . Speeds are estimates; how they are worked out.

What to run it on in the US

Fully in memory at 32k context, priced at retailers today.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA Tesla P40Q5_K_Mfits10 tok/s$34528.1
Intel Arc Pro B60Q5_K_Mfits10 tok/s$65015.8
Mac mini M4 32GBQ4_K_Mfits4 tok/s$1,4202.8
AMD Radeon RX 7900 XTXQ5_K_Mfits24 tok/s$1,53315.6
NVIDIA GeForce RTX 3090Q5_K_Mfits26 tok/s$1,60016.4
AMD Radeon AI PRO R9700Q8_0fits12 tok/s$1,7006.8
NVIDIA GeForce RTX 3090 TiQ5_K_Mfits28 tok/s$2,06013.7
NVIDIA GeForce RTX 4090Q5_K_Mfits28 tok/s$2,07313.6
Mac mini M4 Pro 48GBQ8_0fits6 tok/s$2,1492.7
Mac mini M4 Pro 64GBQ8_0fits6 tok/s$2,9002.0
Mac Studio M4 Max 64GBQ8_0fits12 tok/s$3,4003.4
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits5 tok/s$3,4491.3
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits5 tok/s$3,6501.3
Mac Studio M3 Ultra 96GBQ8_0fits18 tok/s$4,3994.0
NVIDIA GeForce RTX 5090Q8_0fits37 tok/s$4,9217.4
Mac Studio M4 Max 128GBQ8_0fits12 tok/s$4,9992.3
ASUS Ascent GX10 128GBQ8_0fits6 tok/s$6,1000.9
NVIDIA DGX Spark 128GBQ8_0fits6 tok/s$7,3990.8
NVIDIA RTX PRO 6000 BlackwellQ8_0fits37 tok/s$7,9974.6
Mac Studio M3 Ultra 256GBQ8_0fits18 tok/s$8,4992.1
Mac Studio M3 Ultra 512GBQ8_0fits18 tok/s$23,0000.8
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits5 tok/s––
Intel Arc B580Q4_K_Moffload5 tok/s$34613.6
NVIDIA GeForce RTX 3060 12GBQ4_K_Moffload5 tok/s$36013.3
Intel Arc A770 16GBQ4_K_Moffload7 tok/s$40018.3
NVIDIA GeForce RTX 3080 10GBQ4_K_Moffload5 tok/s$4699.8
AMD Radeon RX 7600 XTQ4_K_Moffload6 tok/s$48011.7
AMD Radeon RX 9060 XT 16GBQ4_K_Moffload6 tok/s$52011.3
AMD Radeon RX 7800 XTQ4_K_Moffload8 tok/s$55014.4
NVIDIA GeForce RTX 5060 Ti 16GBQ4_K_Moffload8 tok/s$61812.1
NVIDIA GeForce RTX 4070 SuperQ4_K_Moffload5 tok/s$6997.4
NVIDIA GeForce RTX 5070Q4_K_Moffload5 tok/s$7507.2
NVIDIA GeForce RTX 4060 Ti 16GBQ4_K_Moffload6 tok/s$7707.9
AMD Radeon RX 7900 XTQ4_K_Moffload21 tok/s$80026.5
AMD Radeon RX 9070 XTQ4_K_Moffload8 tok/s$8159.7
NVIDIA GeForce RTX 4070 Ti SuperQ4_K_Moffload9 tok/s$9848.9
NVIDIA GeForce RTX 5070 TiQ4_K_Moffload10 tok/s$9959.6
NVIDIA GeForce RTX 4080 SuperQ4_K_Moffload9 tok/s$1,1997.5
NVIDIA GeForce RTX 5080Q4_K_Moffload10 tok/s$1,6006.1

1 other card or machine we price cannot hold it at this context.

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q4_K_Mgguf13.3 GB15.3 GB19.2 GB24.4 GB34.8 GB
Q5_K_Mgguf15.6 GB17.6 GB21.5 GB26.7 GB37.1 GB
Q8_0gguf23.3 GB25.3 GB29.2 GB34.4 GB44.8 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/mistralai/Devstral-Small-2507_gguf:Q4_K_M
    • LM Studiolms get mistralai/Devstral-Small-2507_gguf@q4_k_m
    • llama.cppllama-server -hf mistralai/Devstral-Small-2507_gguf:Q4_K_M
  • Q5_K_M
    • Ollamaollama run hf.co/mistralai/Devstral-Small-2507_gguf:Q5_K_M
    • LM Studiolms get mistralai/Devstral-Small-2507_gguf@q5_k_m
    • llama.cppllama-server -hf mistralai/Devstral-Small-2507_gguf:Q5_K_M
  • Q8_0
    • Ollamaollama run hf.co/mistralai/Devstral-Small-2507_gguf:Q8_0
    • LM Studiolms get mistralai/Devstral-Small-2507_gguf@q8_0
    • llama.cppllama-server -hf mistralai/Devstral-Small-2507_gguf:Q8_0

Notes

Text-only 24B coding agent (vision encoder removed), said to run on a single RTX 4090 or 32GB Mac. vLLM serving requires vLLM >= 0.9.1 and mistral_common >= 1.7.0.

Questions

How much VRAM does Devstral Small 2507 need?
At Q4_K_M, Devstral Small 2507 needs about 15 GB of memory at 8k context and 19 GB at 32k, counting the weights, the KV cache and runtime buffers.
What is the best GPU for running Devstral Small 2507 locally?
The cheapest hardware in the US that runs Devstral Small 2507 fully in memory at 32k context is the AMD Radeon RX 7900 XTX, $1,533 at Amazon, at about 24 tok/s (estimated). The best value is the NVIDIA GeForce RTX 3090 (about 26 tok/s for $1,600). The fastest we price is the NVIDIA GeForce RTX 5090, at about 37 tok/s.
Can I run Devstral Small 2507 on a 16 GB GPU?
Only partly: at 32k context it does not fit in 16 GB, so part of it runs from system RAM and it slows to about 7 tok/s on the Intel Arc A770 16GB (estimated).
Can I run Devstral Small 2507 on a 24 GB GPU?
Yes: at Q5_K_M it fits fully in 24 GB at 32k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 28 tok/s (estimated).
How do I run Devstral Small 2507 locally?
With Ollama: ollama run hf.co/mistralai/Devstral-Small-2507_gguf:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is Devstral Small 2507 released under?
Apache 2.0, which allows commercial use.