Rubric
The score is a weighted opinion, on purpose.
Which question this answers
StackPick answers which tool to use for a job. Cursor or Windsurf for the editor. Lovable or Bolt for the app. A model API, an image generator, a voice tool. The score is about that product. Windsurf the editor is filed here as Devin Desktop.
OpenRouter rankings and Arena answer which model is ahead. OpenRouter ranks models by tokens processed through the OpenRouter API. OpenRouter states that a higher token total shows how much a model is used, not which model is best for a task. Arena, also called LMArena, ranks models by blind human votes. People compare anonymous answers, and only those anonymous votes count toward the public board. Arena fits the votes with a Bradley-Terry model. Both boards are worth reading. They rank models. This directory ranks tools.
The four parts
- Capability40
- Ease25
- Value20
- Momentum15
Capability. How completely a fluent builder can finish the job this lane is for, including hard tasks, not the marketing demo. For a model API, Arena’s blind votes are one input.
Ease. Time to a useful result, and how little setup or prompt ceremony it takes.
Value. What you get before a sales call. Open-source tools that are free to run score high even when the model API is extra. Unknown dollar prices do not get a fake discount.
Momentum. Whether the product is still how builders work, not a dormant brand with a familiar logo. For a model API, token use on OpenRouter rankings is one input.
External model signals
On the LLM APIs lane, the two boards inform capability and momentum. The rubric still produces the score. Ease and value come from the product and the plan. The tables stay on their sites.
Capability, from Arena. Arena, also called LMArena, ranks models by blind human votes. People compare anonymous answers, and only those anonymous votes count toward the public board. Arena fits the votes with a Bradley-Terry model. When we set capability for an API, we look at how the models that API serves are faring with those voters, then still judge the product around them: docs, SDKs, modalities, and whether a fluent builder can finish the job. Ease and price stay ours. The ratings stay on Arena. LMArena leaderboard. Arena FAQ.
Momentum, from OpenRouter rankings. OpenRouter ranks models by tokens processed through the OpenRouter API. OpenRouter states that a higher token total shows how much a model is used, not which model is best for a task. Sustained use of an API’s models supports momentum for that API, and for OpenRouter as the router people actually send traffic through. Token rank does not set capability. OpenRouter the product is also a tool on this list. Its rankings page is a usage signal, separate from our score of the router. https://openrouter.ai/rankings.
Checked 14 Oct 2026. The OpenRouter rankings page showed usage through 13 Oct 2026. Rankings data is licensed CC BY 4.0. The figures move, so StackPick links the boards. CC BY 4.0.
A worked example
Cursor is filed as capability 96, ease 92, value 80, momentum 97. That is 96×40 + 92×25 + 80×20 + 97×15 = 9195. Divide by 100 and round: 92. 92 is S-tier.
What a tier means
- S · 90 and above. A default. Start here if you want one tool.
- A · 80 to 89. Excellent, with a real tradeoff.
- B · 70 to 79. Solid when your constraint matches the tool.
- C · under 70. On the list for a specific situation, not the default.
An empty tier is allowed. We do not drag a tool into C to fill a row, and we do not inflate a tool into S to make a lane look crowned.
Pricing
free means the software itself is free to run. The starting price is $0, and the notes say if you still pay a model provider.freemium means a usable unpaid path exists or has been offered. paid means we are not willing to claim a lasting free tier. The starting price field is check site whenever the number moves faster than this directory.
Dates
releasePrecision is day, month, or year. A year-only date means “public that year,” not 1 January. addedDate is when the file joined StackPick.
Money on the page
affiliateUrl is optional. If it is set, the main button uses it, is marked rel="sponsored", and the official URL stays visible. sponsored: true prints a Sponsored label. Neither field is an input to the formula. Both are off in the seed catalog.
What we do not do
- No aggregate star ratings. There is no review corpus behind them.
- No scraped “starting at $20” unless a human updates the file and changes the validator.
- No rank boost for affiliates.
Questions
Are StackPick scores benchmarks?
No. They are editorial judgments using a published rubric. We do not run hidden evals, and we do not invent user-review stars.
Can a sponsor buy a higher score?
No. A sponsored flag is a label on the page. It does not change the score, the tier, or the sort order. Sponsored placements are off unless a tool file sets sponsored to true.
What does “check site” mean?
We will not print a dollar price we are not willing to defend today. The plan shape (free, freemium, or paid) is filled in. The number is check site.
How often does this change?
When a tool file is edited. The seed catalog was added on 14 Oct 2026. There is no nightly crawler.
How is StackPick different from OpenRouter rankings or Arena?
StackPick answers which tool to use for a job, such as Cursor or Windsurf, or Lovable or Bolt. OpenRouter ranks models by tokens processed through the OpenRouter API. OpenRouter states that a higher token total shows how much a model is used, not which model is best for a task. That usage informs momentum for an API tool (https://openrouter.ai/rankings). Arena, also called LMArena, ranks models by blind human votes. People compare anonymous answers, and only those anonymous votes count toward the public board. Arena fits the votes with a Bradley-Terry model. Those votes inform capability (https://arena.ai). Ease, value, and the final number stay on this rubric. We link the boards and do not copy their figures.