Score
Average score per attempt row (0–1). Rewards partial credit.
Formula: Mean of all attempt scores across all results rows: SUM(score) / COUNT(*) over the results table.
Ranks models that make consistent partial progress on hard tasks. A model that scores 0.5 on every task beats one that passes half and fails the rest on this metric.
Cost / run
Average total LLM cost per benchmark run in USD.
Formula: SUM(cost_usd) / run_count across all runs for this model.
Use to compare operating cost across models with similar pass rates. Does not account for quality. Combine with $/Pass for a cost-efficiency view.
Latency p50
Median per-task wall time (LLM call + compile + test), in milliseconds.
Formula: 50th percentile of per-task duration_ms: LLM latency + compile time + test time.
Use p50 for a typical-case latency expectation. Unaffected by outlier slow tasks.
Pass Rate
Fraction of distinct tasks solved in any attempt across all runs.
Formula: (tasks_passed_attempt_1 + tasks_passed_attempt_2_only) / tasks_attempted_distinct
Primary ranking metric. Compare models here first: it directly measures how often the model delivers working code.
pass^n (strict)
Fraction of tasks the model solved in every single run (strict consistency).
Formula: tasks where ALL runs produced a passing result / tasks_attempted_distinct
Measures reliability under repetition. High pass^n means the model is unlikely to regress on a re-run, important for CI and production use.
$/Pass
Total cost divided by number of distinct tasks passed. Lower is better.
Formula: SUM(cost_usd) / tasks_passed_distinct across all runs.
Best single cost-efficiency metric. Penalises expensive models that pass few tasks and rewards cheap models with high pass rates.
Latency p95
95th-percentile per-task wall time. Captures tail latency.
Formula: 95th percentile of per-task duration_ms across all tasks in all runs.
Use p95 to understand worst-case latency. A low p95 means the model rarely stalls, relevant for automated pipelines with timeouts.
Overview
OpenAI: GPT-5.4 has run on 3 occasions, attempting 64 tasks with an average score of 60.86.
Settings
Generation parameters used across this model's runs. "varies" indicates the value differed between runs.
- Temperature
- varies
- Thinking budget
- varies
- Avg tokens / run
- 122,617
- Consistency
- 89.1%
History
Cost
Failure modes
- AL0104 132 Syntax error, '{' expected view all →
- AL0000 87 App generation failed view all →
- AL0111 42 Semicolon expected. Add a semicolon (;) to terminate the statement. view all →
- AL0360 36 Text literal was not properly terminated. Use the character ' to terminate the literal. view all →
- AL0107 35 Syntax error, identifier expected. Provide a valid name (letters, digits, and underscores only). view all →
- AL0118 26 The name 'CreateSequentialGuid' does not exist in the current context. view all →
- AL0198 26 Expected one of the application object keywords (table, tableextension, page, pageextension, pagecustomization, profile, profileextension, codeunit, report, reportextension, xmlport, query, controladdin, dotnet, enum, enumextension, interface, permissionset, permissionsetextension, entitlement) view all →
- AL0185 17 Interface 'CG Token Provider' is missing view all →
- AL0126 15 No overload for method 'DefaultRequestHeaders' takes 1 arguments. Candidates: built-in method 'DefaultRequestHeaders()' view all →
- AL0132 13 'Status' does not contain a definition for ' ' view all →
Shortcomings
AL concepts OpenAI: GPT-5.4 struggles with. Click a row for description, correct pattern, and observed error codes.
No shortcomings analyzed yet
Shortcomings analysis is on the roadmap. The first analyzer run is scheduled for the P8 release; until then, this section reflects no data.
See methodologyRecent runs
Methodology
Scores are computed per task, averaged across attempts. See the about page for details.