LLM Benchmark

Detailed Eval Results

How well do leading LLMs write SpacetimeDB code? We prompt each model, run the generated code against live modules, and score with automated checks.

Claude Opus 4.8

Typescript · With Guidelines
0.0%Task Pass Rate
0/89Checks Passed
47Total Evals

Pass Rate Over Time

Click a data point to view results from that date.

auth

0% · 8 evals
StatusTask IDChecksRun Time
t_026_auth_identity_check0/17.0s
t_027_private_vs_public_table0/16.0s
t_041_registered_user_gate0/18.3s
t_042_admin_bootstrap0/18.4s
t_043_role_based_access0/111.0s
t_044_ban_list0/18.7s
t_045_rate_limit0/18.7s
t_046_shared_document0/110.4s

basics

0% · 15 evals
StatusTask IDChecksRun Time
t_000_empty_reducers0/17.8s
t_001_basic_tables0/15.3s
t_002_scheduled_table0/17.6s
t_003_struct_in_table0/16.0s
t_004_insert0/25.7s
t_005_update0/37.2s
t_006_delete0/35.3s
t_007_crud0/45.6s
t_008_index_lookup0/38.1s
t_009_init0/47.4s
t_010_connect0/15.6s
t_011_helper_function0/36.0s
t_038_schedule_at_time0/16.4s
t_039_cancel_schedule0/17.9s
t_040_lifecycle_player0/15.8s

data modeling

0% · 6 evals
StatusTask IDChecksRun Time
t_024_event_table0/16.4s
t_025_optional_fields0/17.8s
t_028_cascade_delete0/111.8s
t_029_filter_and_aggregate0/19.6s
t_030_two_table_join0/18.4s
t_031_unique_constraint0/14.8s

queries

0% · 8 evals
StatusTask IDChecksRun Time
t_022_view_basic0/17.6s
t_023_view_per_user0/14.9s
t_032_range_query0/18.7s
t_033_sort_and_limit0/17.9s
t_034_find_first0/15.5s
t_035_select_distinct0/16.3s
t_036_count_without_collect0/17.1s
t_037_multi_column_filter0/18.3s

schema

0% · 10 evals
StatusTask IDChecksRun Time
t_012_spacetime_product_type0/36.3s
t_013_spacetime_sum_type0/37.3s
t_014_elementary_columns0/37.2s
t_015_product_type_columns0/37.3s
t_016_sum_type_columns0/37.6s
t_017_scheduled_columns0/27.9s
t_018_constraints0/37.4s
t_019_many_to_many0/510.9s
t_020_ecs0/711.5s
t_021_multi_column_index0/56.8s