Sulat.com
AI models
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
Pioneer logo

Model details

Devstral Small 2

Devstral Small 2 is a 24-billion-parameter coding model released by Mistral AI as part of the next-generation Devstral 2 family, sitting alongside a larger 123B variant. The model is published under the permissive Apache 2.0 license, making it open-weight and friendly to self-hosting, custom fine-tuning, and on-premise deployment. Because of its compact size, Devstral Small 2 is intended to run on consumer hardware while still delivering agentic code capability, fitting teams that want local control over their development tooling without relying on cloud-only endpoints.

On real-world software engineering benchmarks, Mistral reports that the broader Devstral 2 family achieves 72.2 percent on SWE-bench Verified, establishing a state-of-the-art position among open-weight code agents with a fraction of the parameters used by competing systems, and framing Devstral Small 2 as a cost-efficient option for end-to-end code automation. The release pairs the model with Mistral Vibe, a native open-source command-line agent that lets developers delegate repository tasks directly from the terminal. Practically, this combination targets engineering teams that want an open, locally deployable coding assistant capable of multi-step agent workflows, custom integration, and tuning for specialized codebases.

Pioneerdevstral-small-2devstral

Quick Info

Powered by
Provider
Pioneer
Model key
devstral-small-2
Release date
Dec 9, 2025
Last updated
Dec 9, 2025
Knowledge cutoff
2025-12
Input modalities
Output modalities
Capabilities

Cost

Input token cost
$0.10
Output token cost
$0.30

Limits

Output tokens
131,072 tokens
Context window
256,000 tokens

Latest news about Devstral Small 2

Pioneer

CoverageBenchmark

Artificial Analysis profiled Devstral Small 2, attributing the model to Mistral and listing it as an open-weights release from December 2025. The page reports an Artificial Analysis Intelligence Index score of 8 for the non-reasoning variant, placing it well above the comparable open-weights small-model median of 6, an The analysis documents technical specifications including a 256K-token context window (roughly 384 pages of 12-point Arial text), 24B total parameters, Apache 2.0 licensing, and Hugging Face–hosted weights, with text and image input and text output. Throughput is measured at 131.1 output tokens per second, faster than

Videos about Devstral Small 2

More models around Devstral Small 2