OpenRouter vs Langfuse

OpenRouter

Langfuse

OpenRouter · 15 features

• 300+ Models — Claude, GPT, Gemini, Llama, Mistral, Qwen, DeepSeek, Cohere, Grok + open-source.
• 60+ Providers — Anthropic, OpenAI, Google, Together, Fireworks, Groq, DeepInfra, Replicate, etc.
• Auto Fallback — Automatic retry to backup provider on failure.
• Bring Your Own Key — Use your own provider keys → pay providers directly + no platform fee.
• Credit System — Prepay credits via card, crypto, or bank.
• Data Retention Controls — Opt-out of training/retention per provider.
• Free Models Tier — 25+ models available at $0 (limited rate).
• Prompt Caching — Automatic cache for identical prefixes (provider-dependent).
• Provider Preferences — Pin preferred providers per request or default.
• Rankings & Stats — Public leaderboard of most-used models.
• Regional Routing — Route requests to specific geographic regions.
• Streaming — SSE + partial completions.
• Structured Outputs — JSON-mode + JSON schema across supporting models.
• Tool Use / Function Calling — Unified tool calling across providers.
• Unified OpenAI-Compat API — Same endpoint for every model + provider.

Langfuse · 16 features

• Annotation Queues — Human reviewers rate traces. Unlimited on Pro+.
• Dashboards — Aggregate metrics, cost, quality across projects.
• Datasets — Curate test sets from production traces. Run experiments.
• EU Cloud Region — GDPR-compliant hosting in EU.
• Evaluations — LLM-as-judge, manual scores, custom model-graded evaluators.
• LLM Cost Tracking — Automatic cost calculation per provider/model.
• OpenTelemetry Native — OTel SDK → Langfuse endpoint works out of box.
• Playground — Test prompts + models + variables live.
• Prompt Management — Version, tag, label prompts. Reference from code by label.
• Public API — Full REST API for ingest, query, prompt management.
• Python @observe decorator — One-line decorator to trace any function.
• Self-Hosting — Docker Compose + k8s Helm chart.
• Sessions — Group related traces (conversations, agent runs).
• Tracing — Capture every LLM call, tool call, nested span with inputs/outputs/cost.
• Users Tracking — Segment traces by user ID, track per-user cost.
• Webhooks — Subscribe to trace completion events.