Open model finder 01
What are you building?
Pick the closest task. Recommendations use TokenWatch’s published benchmarks, provider capabilities, pricing, and performance snapshots—not a universal “best model” score.
Choose a use case
Optional constraints Privacy, provider location, priorities, and model type
Provider data is incomplete. We don’t know every provider’s headquarters, so excluding a country only removes providers whose headquarters we know. Speed and uptime aren’t reported for every provider, so “Fastest” and “Most reliable” rank only on the data we have.
Loading model, benchmark, and provider data…
Recommendations 02
Three ways to choose
Quality scores are relative within each workload; the “good enough” pick uses an absolute benchmark floor.
More options 03
Also considered
Other benchmarked models that pass the use-case gates.
Evidence gaps 04
Not enough benchmark data
These options lack the evidence needed for a benchmark-based rank. They are not necessarily poor models.
Where to run it 05
Providers for this model
Provider scores compare only eligible offerings for this model and workload. Missing telemetry is unknown, not zero.
Unverified — missing required metadata
These providers may offer the model, but we can’t confirm a required capability or context limit. They’re not in the ranked picks.
Next steps 06
Take this recommendation with you
Blended prices are USD per million tokens from the published rate snapshot, not an invoice. Model and provider availability, policy, and prices can change.
How the open model finder works
Each use case has an assumed token mix, required capabilities (such as tool calling, structured output, and a minimum context window), benchmark weights, and an absolute quality floor. Only open-weight models are considered unless you include proprietary models. Best quality is the highest benchmark composite; best value sits on the quality-versus-price Pareto frontier rather than a score-per-dollar ratio; cheapest good-enough is the cheapest model that clears the floor. Providers are ranked on blended price, time to first token, throughput, and uptime, and a close call means the top two are within three points. Creative writing is led by LMArena preference ratings (CC BY 4.0); chat also shows a people’s favourite.
Jump straight to a use case:
- Coding agent
- Tool-using agent
- Chat over my documents
- Extract data to JSON
- Cheap bulk processing
- Chatbot
- Creative writing
- Reasoning / math
- Websites / UI
Read how recommendations work, the model-choice FAQ, or query the same engine through the recommendation API (/api/v1/recommend, /api/v1/recommend/providers, /api/v1/use-cases).