aicharts plots published AI benchmark scores against cost and tokens per task, and a local collector measures your own agents' token use.
By Hraness · 3 min read
Drafted with AI from the source code and reviewed by Codex.
Each result on aicharts keeps its own source, version, and retrieval date, and each benchmark stays on its own scale. Made with SlopCamera
Status: the benchmark charts and notes are free and live at aicharts.io. The usage collector is in development.
The launch film: illustrations of aicharts's charts and usage dashboard, with captions.
See which AI model wins at each price
aicharts puts published AI benchmark scores and the cost of a task on one chart, so you can see which model gives the most for your budget. Every point names its source and the day it was checked.
The homepage chart plots 108 model settings by Artificial Analysis Intelligence Index v4.3.2 score against cost per task. The line joins the models nothing cheaper beats, so the best pick at each price stands out.
The coding chart compares 31 coding-agent setups. Each point is a model, the agent app running it, and an effort setting. Hover one to see its scores, cost per task, time, and tokens.
The benchmarks library covers 62 tests across reasoning, research, memory, images, video and audio. Each one is labelled as a chart, a guide to the source, or an early test, and it keeps its own scale.
The aicharts collector reads the usage files your coding agents already keep, from 55 supported sources, and adds up tokens, cost and speed per model and day. Prompts and transcripts stay on your machine.
A local report is a file you open in your browser tab. On a Mac you can also sync daily totals to your dashboard: token counts, cost and time per agent and model. Prompts, transcripts, file paths and keys stay on your machine.
aicharts collector
#A local report: a file on your machine, opened in your browser
For picking a model on cost, not for one overall rank
aicharts is for choosing a model or coding agent by weighing score against cost, time or tokens. If you want one overall ranking of every model, look elsewhere: aicharts builds no score of its own, and it cannot test your codebase.
aicharts.example
aichartsIntelligence Index v4.3.2 against output tokens
The goal is a catalog where any benchmark result you might use to pick a model shows its setup, version, cost and date. Tests that have only a source guide today are meant to become charts once their data can be checked.
aicharts.example/data
ChartedTerminal-Bench 4JSON ↓
Question
Which agent can finish difficult terminal work?
Measures
Task success over 66 tasks and five trials per configuration, with source-reported 95% confidence intervals.
Source
Harbor Framework
Version
4.0.0
Compare
aicharts’ primary terminal benchmark. Compare exact version 4.0.0 with the named model, agent version, and effort.
Limits
An agent and model are evaluated together; this is not a model-only ranking.
The scores come from their owners, not from aicharts
Scores, costs and token counts come from the benchmark owners and trackers aicharts cites; it runs no tests itself. Costs keep the source's unit, per task or per full run, so compare two only when the unit matches.
The charts are free and live; the collector is early
The charts and notes are free at aicharts.io. Usage collector status: In development. Version 0.4.0 for Linux is on GitHub Releases; on a Mac you build it from source.
aicharts tui
$aicharts tui --snapshot
== Status ==
aicharts
● Collecting · Last pass 4 min ago · last sync 50 min ago
The data page: every entry's source, version and limits. The homepage uses Intelligence Index v4.3.2; the earlier v4.1.1 results stay a separate dataset, and the owners' Terminal-Bench 4.0 results stay apart from Artificial Analysis's own runs.
Sources
Coding AgentsArtificial Analysis, 2026. The public coding-agents comparison is the source of the aicharts coding-agent snapshot. Model names, agent harnesses, settings, AA Index scores, and mean API costs are Artificial Analysis measurements.
LLM LeaderboardArtificial Analysis, 2026. The public models leaderboard is the source of the aicharts Intelligence Index snapshot. Scores, per-task costs, and output tokens are Artificial Analysis measurements under Intelligence Index v4.3.2.
Terminal-BenchHarbor Framework, 2026. The benchmark owners’ repository publishes Terminal-Bench and its versioned task releases. The version-pinned Terminal-Bench 4.0 cohort in the aicharts benchmarks library comes from the owners, separately from Artificial Analysis’s own Terminal-Bench 4 runs.
aicharts
