Devstral-2-123B-Instruct-2512
mistralai/Devstral-2-123B-Instruct-2512
Official · FP8 · safetensors
Runtime certified: vLLM
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| FP8 | safetensors | 8 | Official | Runtime certified |
Mistral · Devstral · Devstral 2
Devstral 2 123B Instruct 2512: Specs, GPU requirements, hosting & deployment
Devstral 2 123B Instruct 2512 is a current Devstral release in the Mistral family, certified from its immutable official source revision.
Total parameters123.0B
Active parametersDense model
Context window262,144 tokens
Release statusCurrent · Current
ArchitectureMinistral3ForCausalLM
ModalitiesText, Code
LicenseModified MIT License
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed With Conditions
Release date2025-11-28
Architecture, strengths and use cases
Devstral 2 123B Instruct 2512 is an official self-hostable Mistral AI release in the Mistral Devstral track. It uses Ministral3ForCausalLM, carries 123B, supports a verified context window of 262,144 tokens, and exposes only source-certified checkpoints and runtime combinations. Its release, license and deployment facts are tied to the official repository revision rather than inferred from the family name.
Devstral 2 123B Instruct 2512 uses Ministral3ForCausalLM with 123B. The stored architecture profile records 88 layers, hidden width 12288, 96 attention heads and a 262,144 token context ceiling where the official configuration publishes those fields.
Verified Ministral3ForCausalLM architecture
262,144 token certified context
Official FP8 checkpoint coverage
Private LLM inference
Agentic workflows
Long-context document processing
Software engineering and code generation
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed vllm runtime configuration with the exact checkpoint precision. The recommendation engine checks weight memory, KV cache, runtime overhead, safety headroom, GPU architecture and topology before ranking rentable and reference hardware.
Estimated weight memory117.99 GB
KV cache1.38 GB
Runtime overhead38.9 GB
Total with safety margin196.25 GB
Recommended best value
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
85.75 GB headroom · Limited capacity
USD 4,690 / monthLowest-cost viable
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
85.75 GB headroom · Limited capacity
USD 4,690 / monthHighest performance
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
85.75 GB headroom · Limited capacity
USD 4,690 / monthLicense and self-hosting considerations
The full 123B resident weight set must fit across the selected GPUs. Context and concurrency add KV-cache memory; MoE active parameters do not replace resident-weight memory.
Devstral 2 123B Instruct 2512 is published under Modified MIT License. The release-scoped official license link and verification date are retained with the catalogue record.
Commercial Use Allowed With Conditions. Commercial use is conditional. Review the linked official terms before deployment.
Alternatives and sibling variants
Devstral Small 2 24B
Release timeline
mistral-ai/mistral/devstral-2-123b-instruct-2512 @ 17e746573df3af5648277fa5ccf3abaa705c7332b91b696e8a9e8347ce21cf2e
Frequently asked questions
How much VRAM does Devstral 2 123B Instruct 2512 require?
The exact estimate depends on checkpoint precision, context and concurrency; the page recommendation uses the certified memory engine and compatible runtime configurations.
Which runtimes support Devstral 2 123B Instruct 2512?
Certified upstream runtime evidence is available for vllm.
Can Devstral 2 123B Instruct 2512 be used commercially?
The stored commercial classification is Commercial Use Allowed With Conditions. Consult the official Modified MIT License terms for the final obligations.
Official references
Last verified: 2026-08-26