Code Models
Code models are trained specifically on code and code-adjacent text to write, complete, explain, or refactor software — some are standalone releases, others are the specific checkpoints powering the AI coding assistants tracked in the companies taxonomy above. This category is judged heavily on repository-scale benchmarks (not just single-function completion) as coding agents increasingly need to reason across a whole codebase rather than one file. GROUNDING tracks new code-model releases and head-to-head coding benchmark results, since this is one of the fastest-moving and most directly builder-relevant model categories.
At a glance
- 7 tracked models (1 hand-curated)
- Most recently updated: qwen2.5-coder-14b (2026-06-25)
Most actively covered
- Claude Code
- Kimi-K2.7-Code ★
- Qwen2.5-coder
- Qwen3 Coder
- Qwen2.5-Coder-3B
- qwen2.5-coder-14b
- qwen2.5-coder-7b
FAQ
What is the Code Models category?
Models specialized for writing, completing, or reasoning about code.
How many models are in this category?
GROUNDING currently tracks 7 models in Code Models, 1 of which have a hand-curated profile.
How current is this hub?
The most recently updated entry is qwen2.5-coder-14b, last mentioned 2026-06-25; the radar refreshes hourly.