# ParetoNet: the live Pareto frontier of LLMs

> Which large language models nothing else beats on intelligence, cost and speed at once. ParetoNet places every model, at each reasoning-effort level, by Artificial Analysis Intelligence Index, cost per task and end-to-end time in an interactive 3D chart at https://paretonet.dev/, updated at least hourly.

Source: [Artificial Analysis (artificialanalysis.ai)](https://artificialanalysis.ai/leaderboards/models), data as of 2026-11-08 11:00 UTC, Intelligence Index v5.

## Pareto-optimal on all three

20 of the 44 models nearest the Pareto frontier, which always include every Pareto-optimal model. No other model is at least as good on all three axes and better on one. Listed most intelligent first; none ranks above another, since each is best for some trade-off.

- **Gemini 4 Carbon · high** (Google): Intelligence 60.0 · Cost / task $7.75 · E2E time / task 31m 03s
- **Claude Opus 5.5 · max** (Anthropic): Intelligence 59.1 · Cost / task $10.04 · E2E time / task 25m 20s
- **Gemini 4 Carbon · medium** (Google): Intelligence 58.3 · Cost / task $6.59 · E2E time / task 25m 55s
- **Claude Opus 5.5 · xhigh** (Anthropic): Intelligence 57.4 · Cost / task $6.05 · E2E time / task 15m 58s
- **GPT-6 Astra · xhigh** (OpenAI): Intelligence 55.3 · Cost / task $3.66 · E2E time / task 10m 44s
- **GPT-6 Astra · high** (OpenAI): Intelligence 54.8 · Cost / task $3.24 · E2E time / task 9m 24s
- **GPT-6.1 Sol · max** (OpenAI): Intelligence 54.6 · Cost / task $1.27 · E2E time / task 20m 25s
- **GPT-6.1 Sol · xhigh** (OpenAI): Intelligence 52.5 · Cost / task $0.76 · E2E time / task 9m 20s
- **GPT-6 Astra · medium** (OpenAI): Intelligence 51.6 · Cost / task $2.73 · E2E time / task 6m 35s
- **GPT-6.1 Sol · high** (OpenAI): Intelligence 50.3 · Cost / task $0.59 · E2E time / task 6m 29s
- **GPT-6.1 Sol · medium** (OpenAI): Intelligence 46.3 · Cost / task $0.36 · E2E time / task 3m 47s
- **GPT-6 Astra · low** (OpenAI): Intelligence 43.3 · Cost / task $1.42 · E2E time / task 3m 32s
- **GPT-6.1 Terra · high** (OpenAI): Intelligence 42.4 · Cost / task $0.26 · E2E time / task 4m 03s
- **GPT-6.1 Terra · medium** (OpenAI): Intelligence 38.2 · Cost / task $0.17 · E2E time / task 2m 37s
- **GPT-6.1 Sol · low** (OpenAI): Intelligence 34.8 · Cost / task $0.16 · E2E time / task 1m 46s
- **Claude Haiku 5.5 · high** (Anthropic): Intelligence 29.9 · Cost / task $0.16 · E2E time / task 6m 30s
- **GPT-6.1 Terra · low** (OpenAI): Intelligence 29.4 · Cost / task $0.07 · E2E time / task 1m 05s
- **Claude Haiku 5.5 · low** (Anthropic): Intelligence 18.6 · Cost / task $0.04 · E2E time / task 2m 11s
- **GPT-6 Luna · medium** (OpenAI): Intelligence 18.5 · Cost / task $0.04 · E2E time / task 3m 24s
- **GPT-6 Luna · low** (OpenAI): Intelligence 9.7 · Cost / task $0.0051 · E2E time / task 36 s

## How to read the chart

- **Pareto-optimal**: no other model shown is at least as good on all three axes and better on one. Ringed marks are Pareto-optimal; the net joins them.
- **Ideal corner**: the cheapest, fastest and most intelligent point of the chart.
- **Nearest 25% and 50%**: every Pareto-optimal model, then the models closest behind the frontier, up to a quarter or half of all measured models. "All" shows every measured model.
- **The list**: every Pareto-optimal model shown, most intelligent first. None ranks above another: each is best for some trade-off.
- **2D views**: draw the frontier for just the two axes on screen, so a model can sit on that staircase without being Pareto-optimal on all three.
- **Artificial Analysis Intelligence Index v5**: higher is better. Scores are not comparable across index versions.
- **Cost per task**: average cost of one Intelligence Index task at list prices. Log scale, lower is better.
- **E2E time per task**: Artificial Analysis’s published time per Intelligence Index task: task output tokens divided by end-to-end output speed. Models missing the required timing data are omitted from every view. Log scale, lower is better.
- **Labs**: click a lab to hide it; double-click or shift-click one (or a model in the chart) to show only that lab.
- **Dashed outlines**: new models still missing one measurement; they appear only in the 2D view that does not need it.

---

Data from [Artificial Analysis](https://artificialanalysis.ai/leaderboards/models), updated at least hourly. ParetoNet is independent and not endorsed by Artificial Analysis.