LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Devstral 2
◆ OPEN
by Mistral
OPEN WEIGHTSLOCALAPILONG CONTEXTPAIDCLOSED
RANKED #6 OF 10 CODE · OVERALL #226 OF 6,587 · VIBE SCORE 5.7 · 51 VOICES
Devstral 2 is a state-of-the-art open-source model by Mistral AI specializing in agentic coding. It is a 123B-parameter dense transformer model supporting a 256K context window. Devstral 2 supports exploring...
HAPPY9.2
9 mentions
SAD5.8
4 mentions
VIBE TRAJECTORY
26 weeks · 39 voices
HOLDING UP
Recent vibe is steady. No clear fade in the crowd.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Devstral 2 once and it's yours. No subscription, no rate limits, works offline.
123B parameters dense, a Q4_K_M GGUF is about 72 GB
- BARE MINIMUM
Apple Mac Studio M2 Max with 96 GB Unified Memory
96 GB UNIFIED MEMORY
~$2200 USEDDevstral 2 (123B) is too large for single consumer GPUs. Runs Q4_K_M quant in unified memory with ~16k context
- SWEET SPOT
Apple Mac Studio M2 Ultra with 128 GB Unified Memory
128 GB UNIFIED MEMORY
~$3400 USEDQ4_K_M or Q5_K_M quant with 64k to 128k context comfortably in local MLX or llama.cpp
- FULL POWER
Custom PC with 4x NVIDIA RTX 3090 24GB
96 GB VRAM
~$3200 USEDEXL2 4.5bpw or Q4 quantized split across GPUs via vLLM for high throughput and long context
LEARN HOW TO SET IT UP
WILL MY BOX RUN DEVSTRAL 2?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Devstral 2 takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMApple Mac Studio M2 Max with 96 GB Unified Memory | SWEET SPOTApple Mac Studio M2 Ultra with 128 GB Unified Memory | FULL POWERCustom PC with 4x NVIDIA RTX 3090 24GB |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~45 S | ~23 S | ~13 S |
| Summarize a documenta long report boiled down to the points that matter | ~2.3 MIN | ~68 S | ~39 S |
| Build a websitea small landing page, markup and styles together | ~12 MIN★★★★★ | ~6.1 MIN★★★★★ | ~3.5 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~38 MIN★★★★★ | ~19 MIN★★★★★ | ~11 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~23 MIN★★★★★ | ~11 MIN★★★★★ | ~6.6 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
Not enough discussion about Devstral 2 yet to call it. We found 51 posts but people did not say much either way.
41 POSTS · 10 COMMENTS · REDDIT · GITHUB · LEMMY
PEOPLE LOVE
- No fan mail yet. The crowd is still warming up.
PEOPLE COMPLAIN ABOUT
- No complaints yet. The internet is still looking for one. Suspicious.
LOOKING FOR AN ALTERNATIVE?
Other models the crowd has fully reviewed, starting with code models like this one.
Qwen3 Coder 480B A35B
Grok Code
Deepseek Coder> SOURCES (12)
- REDDIT · 2026-01-25 · 385 UPVOTESOllama Models Ranked by VRAM Requirements
Ollama Models Ranked by VRAM Requirements
- REDDIT · 2025-12-16 · 120 UPVOTESLabs - Mistral Small Creative
Labs - Mistral Small Creative
- REDDIT · 2026-02-01 · 78 UPVOTESI love Mistral
I love Mistral
- REDDIT · 2025-12-16 · 70 UPVOTESMy Local coding agent worked 2 hours unsupervised and here is my setup
My Local coding agent worked 2 hours unsupervised and here is my setup
- REDDIT · 2025-12-17 · 35 UPVOTESWe ran 34 models on fresh SWE GitHub PR tasks (November 2025), and GPT-5.2 matches Claude Sonnet 4.5 while being about 2.7x cheaper
We ran 34 models on fresh SWE GitHub PR tasks (November 2025), and GPT-5.2 matches Claude Sonnet 4.5 while being about 2.7x cheaper
- REDDIT · 2025-12-12 · 34 UPVOTESEurope's devstral-small-2, available in the ollama library, looks promising
Europe's devstral-small-2, available in the ollama library, looks promising
- REDDIT · 2026-02-19 · 25 UPVOTESGLM-5 is officially on NVIDIA NIM, and you can now use it to power Claude Code for FREE 🚀
GLM-5 is officially on NVIDIA NIM, and you can now use it to power Claude Code for FREE 🚀
- REDDIT · 2026-04-30 · 24 UPVOTESMaybe it would be worth focusing on small, dense models for developers and not only?
Maybe it would be worth focusing on small, dense models for developers and not only?
- REDDIT · 2026-04-09 · 24 UPVOTESWhen are we gonna get more 1-Bit models(Medium & Large size)?
When are we gonna get more 1-Bit models(Medium & Large size)?
- REDDIT · 2026-05-29 · 21 UPVOTESI run ollamatps.com to track TPS — after my $20 Ollama Cloud Pro plan expired, I tested the free tier and ranked all 25 models by real…
I run ollamatps.com to track TPS — after my $20 Ollama Cloud Pro plan expired, I tested the free tier and ranked all 25 models by real quota cost (same request to each)
- REDDIT · 2026-03-11 · 12 UPVOTESMistral, scammers, false advertising, Follow up to my usage on vibe
Mistral, scammers, false advertising, Follow up to my usage on vibe
- REDDIT · 2026-06-08 · 0 UPVOTESWhy do people use Ollama?
Why do people use Ollama?