Explore what independent evaluations measure, which models they cover, and where the results apply.
Artificial Analysis
Artificial Analysis · Text-to-video models. What it measures: Published evaluation protocol. Read the publisher’s evaluation method, task scope, and tested model version before applying its results to a tool. No tool score is inferred here.
VBench team
VBench team · Text-to-video models. What it measures: Quality and consistency dimensions. A multi-dimensional video benchmark covering visual quality, motion and consistency.
Artificial Analysis
Artificial Analysis · Text-to-speech providers. What it measures: Published evaluation protocol. Read the publisher’s evaluation method, task scope, and tested model version before applying its results to a tool. No tool score is inferred here.
TTS-AGI & contributors
TTS-AGI & contributors · Text-to-speech models. What it measures: Published evaluation protocol. Read the publisher’s evaluation method, task scope, and tested model version before applying its results to a tool. No tool score is inferred here.
BigCode project
BigCode project · Code-generating models, not editors. What it measures: Practical coding task completion. Measures model performance on multi-library programming tasks. It is useful context, but it does not rank Cursor, Windsurf or Bolt as products.
Artificial Analysis
Artificial Analysis · Text-to-image models. What it measures: Published evaluation protocol. Read the publisher’s evaluation method, task scope, and tested model version before applying its results to a tool. No tool score is inferred here.
Hugging Face & contributors
Hugging Face & contributors · Speech-recognition models only. What it measures: Transcription accuracy and speed. Plots word error rate against speed across ASR models. It does not measure meeting summaries, action items or integrations.
Datacurve
Datacurve · Coding agents run through mini-swe-agent. What it measures: Long-horizon engineering task success. Measures frontier coding agents on 113 original, long-horizon engineering tasks across 91 repositories and five languages.
