Best local AI model per tier for long coding tasks
Artificial Analysis Intelligence Index
| Tier | Intelligence Index | Best model | Params (B) | Size on disk (4-bit / full) |
|---|---|---|---|---|
| Loading data… | ||||
The Intelligence Index is a general index across all of Artificial Analysis's evaluations, shown so every model is rated on one scale. It is not the basis of the tiers: it rates Nanbeige4.3-33B and Qwen3.8 27B almost equally, although Nanbeige is about twice as good on long coding tasks.
All models, share of Sonnet 5.5
Long-coding score (average of Terminal-Bench 4.0 and FrontierCode 2.0) as a percentage of Sonnet 5.5's 52.7, for every model in the data file. Dashed lines are tier edges. Open-weights models are coloured by tier; cloud-only models are grey.