We publish two distinct views: task-fit scores for the whole catalog, and original results for models included in named benchmarks. Task-fit scores count documented features using the weights below. The weights are Aivorix editorial choices; provider documentation supports the feature inputs. We have not run a laboratory evaluation or reproduced a third-party review score.

How other ranking sites score products

G2 combines user satisfaction and market presence within product categories. Its review-based score is different from a coding or reasoning test.

Artificial Analysis combines independently run evaluations with use-case weights. Aivorix does not claim to reproduce that methodology or its results through feature counting.

Our task-fit formula

Score = sum of weights for criteria supported by the linked provider evidence. Each criterion receives its full weight or zero; weights total 100 in every category. The same rubric applies to every catalog entry. We do not normalize to the current leader or add a popularity bonus.

A zero means no supporting feature evidence has been recorded for that criterion. It can reflect a specialist product or a gap in our review; it does not prove the product is incapable. A score of 100 means all five criteria are documented, not perfect quality. Equal scores share a competition rank (1, 1, 3); tied entries are displayed alphabetically.

AI workflow coverage

  • Writing / text responses: 20 points
  • Document or knowledge-source input: 20 points
  • Image / video creation: 20 points
  • Voice / audio workflow: 20 points
  • Tool calls / agent actions: 20 points

General-work task fit

  • Explicit reasoning / analysis workflow: 30 points
  • Writing / text responses: 15 points
  • Document or knowledge-source input: 20 points
  • Image input / editing: 15 points
  • Code generation / editing: 20 points

Research workflow fit

  • Web search: 25 points
  • Traceable source citations: 25 points
  • Document or knowledge-source input: 20 points
  • Dedicated multi-step research: 20 points
  • Writing / text responses: 10 points

Tool and integration coverage

  • Tool calls / agent actions: 35 points
  • External app / MCP connections: 25 points
  • Hosted code or terminal execution: 20 points
  • Web search: 10 points
  • Document or knowledge-source input: 10 points

Coding workflow fit

  • Code generation / editing: 30 points
  • Repository coding workflow: 25 points
  • Hosted code or terminal execution: 25 points
  • Tool calls / agent actions: 10 points
  • Document or knowledge-source input: 10 points

Evidence boundaries

Every homepage row links the reviewed source and states its product surface. An API model may require application integration; a product feature may require a paid plan. Code execution means a documented hosted runtime or terminal, not merely an API caller running its own function. Repository support means a documented coding workflow across repository files, not a generic file upload.

Provider feature descriptions are not hands-on tests. Citation support does not guarantee accurate citations, and the presence of a reasoning mode does not establish reasoning quality. Review scope and feature support were checked September 7, 2026. Source publication dates can be older.

Published benchmarks

The separate benchmark view uses OpenAI's Astra evaluation tables, with model names, evaluation versions and provider-report attribution. AA Intelligence Index, BrowseComp, OSWorld and DeepSWE have different tasks and units; their values are never averaged into the task-fit score. Only models with reported results appear in that view. Different provider harnesses, model effort and production app settings can change outcomes.