AI Model Release Tracker
Supabase coding

Supabase Evalswith skills

Supabase's own open benchmark: a coding agent is dropped into a real Supabase project and asked to do real work — set up a schema, fix a broken security policy, debug an Edge Function — and every run is checked against a live Supabase stack. This is the headline number, where the agent has Supabase's own skills loaded, as most people building on Supabase would. The score is the share of scenarios it got right. Higher is better.

Scores from Supabase Evals, which runs the benchmark and publishes the full field.

with skillsno skills

Rankings

Higher is better
#1
Claude Sonnet 5
91.3%
#1
Claude Opus 5
91.3%
3
GPT-5.6 Sol
89.9%
4
Kimi K3
84.1%
5
GPT-5.6 Luna
68.1%
6
GPT-5.4 mini
65.2%

Supabase Evals — frequently asked questions

Supabase's own open benchmark: a coding agent is dropped into a real Supabase project and asked to do real work — set up a schema, fix a broken security policy, debug an Edge Function — and every run is checked against a live Supabase stack. This is the headline number, where the agent has Supabase's own skills loaded, as most people building on Supabase would. The score is the share of scenarios it got right. Higher is better.