XNews

Open‑source model adoption trends

Open‑source model adoption trends

Reports highlight open-source model usage patterns and new model releases. - OpenRouter usage report: >50% open-source model use is roleplay/story/interactive fiction; programming accounts for ~20%. - Geographic shifts: Asia share of open-source usage rose from 13% to 31% in one year. - "First-solution wins": users stick to the first model that solved their pain points even if later models improve. - Middle-sized models are rising in popularity versus tiny or very large models. - Inference model usage rose from near 0% to >50% during 2025. - Mistral released Devstral Small 2: a 24B-parameter open model claimable to run locally on laptops and compete on code tasks with larger models.

Key insights

Roleplay and interactive uses dominate open-source model workloads

Usage data indicates over 50% of open-source model activity is for roleplay, virtual characters, interactive stories, and text-based games, while programming-related usage represents roughly 20% of open-source model use. The pattern reflects permissive licensing attracting conversational and creative usage.

Regional adoption of open-source models is growing in Asia

Open-source model usage in Asia reportedly rose from about 13% to 31% within a year, with several countries (China, Singapore, Korea, Japan) contributing to the increase—indicating rapid geographic diffusion.

Mid-sized models and inference-mode workloads are expanding

The ecosystem is shifting away from many tiny models and toward mid-sized models that balance capability and cost. Inference-mode model use climbed from near 0% in early 2024 to over 50% in 2025, reflecting more multi-step, stateful usage patterns.

Mistral's Devstral Small 2 targets local, privacy-preserving coding workflows

Mistral released Devstral Small 2, a 24B-parameter open model designed to run on local machines and claimed to offer competitive code performance compared with larger models—promoting privacy and on-device development workflows.

Source posts