The whole dataset, free to download

432 models and 303 model and host pairs: price, cached price, context, precision, uptime and quality indices. The per host file is the part you will not find elsewhere. It is what shows the same weights selling at different prices.

CC BY 4.0. Use it commercially, redistribute it, build on it. The licence asks only that you credit BestHunt. Refreshed daily.

models.csvCSV
One row per model, 31 columns including arena Elo
models.jsonJSON
Same data with a metadata envelope
hosts.csvCSV
One row per model and host, with price and precision by provider
hosts.jsonJSON
Host rows with the metadata envelope
categories.csvCSV
One row per model and category, with Elo per task joined to price
categories.jsonJSON
Category rows with the metadata envelope

Fields

id
Model identifier, vendor/model form
aa_intelligence, aa_coding, aa_agentic
Artificial Analysis indices, null where unscored
price_in_usd_per_1m / price_out_usd_per_1m
Median across hosts, US dollars per million tokens
price_cached_in_usd_per_1m
Cache-read rate where the vendor publishes one
price_min_host / price_max_host / price_spread_x
Cheapest and dearest host, and the ratio
context_typical
Context most hosts serve, not the advertised maximum
context_min_host / context_max_host
Range across hosts
best_precision / low_precision_somewhere
Highest precision offered, and whether any host runs 4-bit
provider_count, uptime_24h_pct
Number of hosts and median 24-hour uptime
reasoning_mandatory
True where reasoning tokens are always emitted and billed as output

Null means the value is genuinely unknown, never estimated. Prices are US dollars per million tokens. See the findings for how each figure is derived.

Citing it

Building something on this? The embeddable chart stays current on its own, and the JSON files carry permissive CORS headers if you would rather fetch them directly.