The right model,for every task.
PathFinder classifies your request, runs it through a continuous benchmark ranking, and routes to the model with the best score for your task — with your filters applied. Drop-in compatible. Zero config.
- 127+
- models benchmarked
- 6
- routing filters
- 76%
- avg cost saving
$ curl api.pathfinder.dev/v1/chat \
-d '{"messages": [...]}'
# ✓ Task detected: code-generation
# ✓ Top model: deepseek-coder-v2 (94.2)
# ✓ Filter: cheapest in top-3
# ✓ Routed to: deepseek-coder-v2-eu
# → 76% cost savingOne request.
The perfect model.
Drop in your request
PathFinder uses the standard Chat Completions API format — no SDK changes, no migration. Optionally hint the task type, or let PathFinder detect it from your message automatically.
An automated evaluator ranks the models
A continuous benchmark pipeline ingests public leaderboard scores, community evaluations, and per-task failure rates to produce a weighted ranking across 127+ models — refreshed as models evolve.
The best model gets your request
Your active filters narrow the ranked list. PathFinder picks the top-qualifying model and routes your request — with full logging so you always see exactly why.
Picking a model manually is a guess.
PathFinder makes it a decision.
Stop guessing which model is best. PathFinder knows.
There are 100+ AI models available today — each better at different tasks, at different prices. Nobody has time to track all of them. PathFinder does it for you, automatically, on every request.
Calling models directly
- You hardcode a model — and hope it fits the task
- Wrong model for the job → unreliable, mediocre output
- Paying top-tier prices for tasks any cheap model handles
- No way to know which of 100+ models is actually best
- Switching models means code changes and testing
Task-aware routing (PathFinder)
- PathFinder detects the task type from your request
- Continuously benchmarked model matched to your task
- Cheapest model in the quality-ranked top-3
- Rankings refresh as new models are released
- Same API format — no code changes needed
The right model for code generation is rarely the right model for creative writing.
PathFinder knows the difference — and routes accordingly.
Every task deserves
its champion model.
A continuous benchmark pipeline evaluates models across task categories using public leaderboards and real-world evaluations. When you send a request, PathFinder routes to the current best — not a model you hardcoded months ago.
Code generation
Complex reasoning
Text summarisation
Creative writing
Data extraction
Q&A / retrieval
Your rules.
Our routing engine.
Composable — stack them freely. PathFinder finds the highest-ranked model that satisfies every constraint you set.
Always routes to the lowest-cost model that meets your quality bar. Save up to 80% vs direct API calls.
Prioritise models with lower energy usage and verified carbon footprints. Route green by default.
Route only through providers that keep inference and data within European infrastructure. Built for GDPR compliance.
Route only through US-based providers for teams with American data residency or sovereignty requirements.
Optimise for speed. Automatically routes to the fastest responding model. Ideal for real-time apps.
Route to the top-performing models on your benchmark of choice. Quality, always on.
Unlock advanced
routing features.
All plans include pay-as-you-go model costs. Platform plans add filters, analytics, and compliance controls.
Model costs passed through at provider rate. No markup.
Starter
- 2 routing filters (cheapest, fastest)
- Pass-through model pricing
- Routing fee: $0.40 / 1,000 req
- 7-day log retention
- 1 workspace · community support
Builder
- All 6 routing filters (eco, EU, US, quality…)
- EU / US data residency (GDPR ready)
- Routing fee: $0.20 / 1,000 req
- 90-day log retention
- 3 workspaces
- Email support
Scale
- Custom model fallback chains
- Routing fee: $0.08 / 1,000 req
- 1-year log retention · unlimited workspaces
- Benchmark API access
- SLA 99.9% · dedicated Slack support
Cost example
1,000 requests · avg. 800 tokens each · cheapest filter active
vs ~$8.00 using GPT-4o directly
All prices exclude model costs. Model costs are billed separately at provider market rates. No markup, ever.
Join the waitlist.
PathFinder routes every AI request to the model that wins for your task. Zero config. Zero guessing.