Back to models
GPT OSS 20b
OpenRouter
131K context$0.03/M input$0.14/M output
1197peak 1308
Avg Score
57.5
Avg Cost
<$0.01
Score/$
6031.5
Runs
65
Win/Loss/Draw
Scoring Dimensions
Score Distribution
Category ELOs
multi-languagehard
1993multi-language
1533code-reviewhard
1433backendmedium
1316debuggingmedium
1311debugging
1274debuggingexpert
1246debugginghard
1245backend
1241backendhard
1228backendeasy
1227code-review
1206frontendmedium
1206full-stackhard
1178full-stack
1110frontend
1107code-reviewmedium
1098backendexpert
1040multi-languageexpert
994from-scratch
953refactoringmedium
872full-stackmedium
867refactoring
854from-scratchexpert
849from-scratchhard
840frontendhard
616from-scratchmedium
461frontendeasy
37from-scratcheasy
0refactoringexpert
0frontendexpert
0All Results
| Task | Category | Score |
|---|---|---|
| Build distributed node cluster with gossip protocol | from-scratch | 33.8 |
| Split 1100-line god file into proper modules | refactoring | 22.9 |
| Find and patch all OWASP Top 10 vulnerabilities | debugging | 56.1 |
| Write integration tests for payment flow | code-review | 74.2 |
| Add streaming SSE endpoint for LLM chat | backend | 82.8 |
| Add Google OAuth2 login to Express app | full-stack | 68.0 |
| Migrate callback-hell Express app to async/await | refactoring | 50.6 |
| Implement background job scheduler with persistence | backend | 0.6 |
| Write Kubernetes manifests for Node.js microservice | full-stack | 61.2 |
| Build REST API from scratch | from-scratch | 63.0 |
| Add i18n with locale routing to Next.js app | full-stack | 35.6 |
| Build LLM evaluation harness with structured grading | backend | 44.0 |
| Find and fix 4 hidden backdoors in Flask app | debugging | 65.3 |
| Fix broken GitHub Actions CI pipeline | debugging | 82.2 |
| Add file upload with S3 presigned URLs | backend | 69.0 |
| Add caching layer to eliminate slow SSR page loads | full-stack | 62.0 |
| Fix race conditions in order matching engine | backend | 67.0 |
| Replace console.log with structured logging | refactoring | 43.3 |
| Fix hallucination and context window bugs in RAG agent | backend | 57.3 |
| Build SaaS admin dashboard from scratch | from-scratch | 36.6 |
| Refactor monolithic handler to CQRS | refactoring | 28.0 |
| Build terminal UI dashboard | from-scratch | 50.5 |
| Add GraphQL layer over REST API | multi-language | 85.0 |
| Fix auth bypass vulnerability | debugging | 34.5 |
| Fix data integrity bugs in denormalized e-commerce schema | debugging | 78.9 |
| Fix Node.js stream backpressure causing OOM on large files | backend | 87.0 |
| Port Python CLI to Rust | multi-language | 48.8 |
| Build RAG pipeline with vector search | backend | 39.0 |
| Fix broken responsive layout | frontend | 52.7 |
| Add Redis caching layer to Express API | backend | 51.0 |
| Harden insecure Docker setup with 12 vulnerabilities | code-review | 64.2 |
| Fix React hydration mismatch | frontend | 59.0 |
| Implement transformer inference engine with KV cache | from-scratch | 66.7 |
| Debug race condition in worker pool | debugging | 80.5 |
| Build materialized view refresh pipeline for analytics | backend | 54.8 |
| Write tests for untested legacy Flask service | code-review | 15.1 |
| Implement JWT auth middleware | backend | 47.5 |
| Add slash commands and moderation to Discord bot | backend | 58.0 |
| Code review: identify security vulns | code-review | 76.3 |
| Add cursor-based pagination to REST API | backend | 80.0 |
| Add rate limiting middleware | backend | 63.9 |
| Implement multi-tenant row-level security in Postgres | backend | 24.2 |
| Implement Stripe webhook handler | backend | 84.3 |
| Implement zero-trust API authentication layer | backend | 58.8 |
| Build MCP server for database management | backend | 70.6 |
| Fix flaky test suite | debugging | 73.0 |
| Debug and fix 6 broken database triggers and constraints | debugging | 79.3 |
| Dockerize Node.js monorepo | full-stack | 53.2 |
| Optimize bloated React bundle under 500KB | frontend | 55.3 |
| Build production website with auth and members area | frontend | 42.3 |
| Fix memory leak in event handler | debugging | 78.9 |
| Add retry logic and dead letter queue to Python task queue | backend | 69.3 |
| Build real-time portfolio risk calculator | backend | 45.9 |
| Add virtual scrolling to table rendering 5000 rows | frontend | 80.3 |
| Build CLI tool with subcommands and config | from-scratch | 39.8 |
| Convert React app to PWA with offline support | frontend | 50.9 |
| Fix N+1 query in dashboard | backend | 52.4 |
| Optimize slow Postgres queries in Flask app | backend | 73.4 |
| Zero-downtime schema migration | full-stack | 58.9 |
| Fix deadlocking transaction patterns in Flask app | backend | 57.7 |
| Fix 12 WCAG accessibility violations in checkout form | frontend | 67.2 |
| Remove AI slop and over-engineering from codebase | refactoring | 66.6 |
| Build codebase indexer for LLM context windows | from-scratch | 37.5 |
| Write complex SQL report with window functions | backend | 47.3 |
| Add WebSocket real-time updates | full-stack | 73.8 |