Compare model performance across 15+ benchmarks with ~400 entries from Artificial Analysis. The Benchmarks tab offers browse and compare modes with head-to-head tables, scatter plots, and radar charts. The CLI provides filtering, sorting, and JSON output.

The Benchmarks tab has two modes:

• Browse mode -- model list on the left, detail panel on the right
• Compare mode -- model list on the left, comparison view on the right (H2H table, scatter plot, or radar chart)

The left panel can toggle between a Models list and a Creators sidebar with t.

Press the number key once to sort by that metric; press again to toggle direction:

The sort picker popup lists all available benchmark metrics. Select one with Enter or dismiss with Esc.

Press t to toggle the left panel between the model list and the creators sidebar. The sidebar shows 40+ model creators with counts, filterable by region, type, and open/closed source. Select a creator to filter the model list to their models only.

In browse mode, the detail panel shows full benchmark data for the selected model:

• Indexes -- Intelligence, Coding, Math, GPQA Diamond
• Scores -- individual benchmark scores
• Performance -- speed (tokens/second), latency, time to first token
• Pricing -- input and output cost per million tokens

Values are formatted as {:.1} (one decimal place) for scores and indexes. Missing values display as an em-dash.

Select up to 8 models for comparison:

A side-by-side comparison table showing all metrics for selected models. Press d to show the detail overlay. Scroll with arrow keys when the compare panel is focused.

A two-axis scatter plot comparing selected models. Cycle the axes:

A multi-axis radar chart overlaying selected models. Press a to cycle through radar presets (different metric combinations).

Each selected model gets a unique color from the comparison palette for consistent identification across all three views and the legend.

Press / to search benchmark entries by name, slug, or creator.

The benchmarks CLI can be invoked as models benchmarks <command> or as a standalone benchmarks <command> via a symlink (see Installation#setting-up-command-aliases).

models benchmarks list
models benchmarks list --sort speed --limit 10
models benchmarks list --creator openai --reasoning
models benchmarks list --open --sort price-input --asc

Opens an inline terminal picker with model table and detail preview. Inside the picker:

• / starts a live text filter over name, slug, and creator
• s cycles sort metrics
• S reverses the current sort
• Enter prints the selected model's benchmark details

models benchmarks show gpt-4o
models benchmarks show "Claude Sonnet 4"

Prints a formatted benchmark breakdown. If the query matches multiple variants in an interactive terminal, the picker reopens with just the matching candidates.

models benchmarks list --creator anthropic --json
models benchmarks show gpt-4o --json

Benchmark data is fetched fresh from the Artificial Analysis CDN on every launch -- there is no local cache for benchmark data. The upstream dataset is updated automatically every 30 minutes via a GitHub Actions workflow.