Skip to content
Local AI Wizard
Your machine
Theme

Qwen3 Coder 30B vs Devstral Small 2 for coding

Two open coding models for a 24 to 48 GB machine

Facts from the reviewed catalog

Facts from the reviewed catalog
SpecificationQwen3 Coder 30B A3BDevstral Small 2 24B
TaskText & chatText & chat
Parameters30.5B · 3.3B active24B
LicenseApache 2.0Apache 2.0
Commercial useAllowedlicense textAllowedlicense text
Quality · LMArena 2026-10-22 · textNot on the LMArena text leaderboard of 2026-10-22 · sourceNot on the LMArena text leaderboard of 2026-10-22 · source
MemoryQ4 planning estimate (~23 GB of weights before the machine's reserve)Q4 planning estimate (~18 GB of weights before the machine's reserve)
Checkpointofficial cardofficial card

Neither checkpoint has a result in LMArena 2026-10-22 · text, so this source cannot compare them; no score is inferred. Blind pairwise votes by people on the hosted model, CC BY 4.0. A rating is relative; overlapping intervals mean the order is not settled. Scale: Arena rating (style control, overall) ± half the confidence interval.

Memory on three machines

Each machine's own reserve is added (graphics cards 2 GB; unified and system memory 25%, 2 to 8 GB). Planning estimates, not measured runs.

Qwen3 Coder 30B vs Devstral Small 2 for coding: memory on three machines
MachineQwen3 Coder 30B A3BDevstral Small 2 24B
RTX 4090 · 24 GB VRAM~25 GBIn memory, outside the fast part · offload to PC RAM (32 GB assumed)~20 GBFast memory holds it
RTX 5090 · 32 GB VRAM~25 GBFast memory holds it~20 GBFast memory holds it
Mac mini M4 Pro · 48 GB unified~31 GBFast memory holds it~26 GBFast memory holds it

Qwen3 Coder 30B A3B: Q4 planning estimate (~23 GB of weights before the machine's reserve). Devstral Small 2 24B: Q4 planning estimate (~18 GB of weights before the machine's reserve). Formulas

Everything that runs on each machine: RTX 4090 · 24 GB VRAM · RTX 5090 · 32 GB VRAM · Mac mini M4 Pro · 48 GB unified

Which to choose

Both are Apache 2.0 coding models with a 262,144-token context. Qwen3 Coder is a mixture of experts (30.5 billion parameters, about 3.3 billion active per token), so it generates quickly once its roughly 23 GB of Q4 weights fit. Devstral Small 2 is dense (24 billion) and also reads images; Mistral states it fits a single RTX 4090 or a 32 GB Mac. Neither has a rating in the LMArena text release. Mistral's card reports coding results; Alibaba's card for this checkpoint states none, so the numbers below are not a head-to-head.

Files and run commands

The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.

Qwen3 Coder 30B A3B

unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF · Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf · Q4_K_M · 18.6 GB · unsloth

ollama run hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M
llama-server -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M
lms get https://huggingface.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF
  • mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac

Devstral Small 2 24B

unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF · Devstral-Small-2-24B-Instruct-2512-Q4_K_M.gguf · Q4_K_M · 14.3 GB · unsloth

ollama run hf.co/unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF:Q4_K_M
llama-server -hf unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF:Q4_K_M
lms get https://huggingface.co/unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF
  • mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac

Numbers each publisher reports

Each card's own results with its own settings. Publishers choose different tests and settings, so these values are never compared across publishers here.

Qwen3 Coder 30B A3B · reported by Alibaba

The publisher's card states no benchmark result for this checkpoint (read 2026-10-25). not published on the card · read 2026-10-25

Devstral Small 2 24B · reported by Mistral

3 results · Mistral's model card · read 2026-10-25
SWE Bench Verified
68.0%
SWE Bench Multilingual
55.7%
Terminal Bench 2
22.5%

Values as printed on the card, with the publisher's own settings (card section: Benchmark Results); column “Devstral Small 2”. Not comparable across publishers and not used in any ranking. Model card

Limits and dates

The published quality benchmark does not test a local quantized build. Memory estimates are not speed measurements or guaranteed hardware fits. Repository size, when available, is not VRAM. Formulas

Page updated · Catalog reviewed · LMArena 2026-10-22 · text