Inadviser

Changelog

What changed between snapshots: verdicts, ranks and prices. Each snapshot is kept as a JSON file with the inputs it was made from.

Which AI model gives the most useful agent work for the least cost per task?

  • No change in verdict, ranks, prices or inputs.

Which AI search model gives the most quality for the money?

  • First snapshot. Verdict: Shortlist. 19 options, 35 evidence rows, 0 tariff rows.

Which AI chat model gives the most quality for the money?

  • First snapshot. Verdict: Shortlist. 21 options, 37 evidence rows, 0 tariff rows.

Which AI image model gives the most quality for the money?

  • First snapshot. Verdict: Shortlist. 19 options, 27 evidence rows, 0 tariff rows.

Which AI video model gives the most quality for the money?

  • First snapshot. Verdict: Shortlist. 16 options, 19 evidence rows, 0 tariff rows.

Which electric or hybrid car costs least to run per 100 km in Spain?

  • First snapshot. Verdict: Insufficient evidence. 30 options, 30 evidence rows, 0 tariff rows.

Which text-to-speech service gives the best Spanish ad voice, with commercial rights, for the least money?

  • No change in verdict, ranks, prices or inputs.

Which text-to-speech service gives the best English narration of a 10-hour book for the least money?

  • No change in verdict, ranks, prices or inputs.

Which text-to-speech service gives acceptable English audio at the lowest cost per million characters for a very large catalogue?

  • No change in verdict, ranks, prices or inputs.

Which text-to-speech service speaks fastest and still sounds natural for a customer-service voice agent, for the least money?

  • No change in verdict, ranks, prices or inputs.

Which text-to-speech service gives the best narration for a monetised English YouTube channel for the least money?

  • No change in verdict, ranks, prices or inputs.

Which AI model gives the most useful agent work for the least cost per task?

  • First snapshot. Verdict: Shortlist. 57 options, 110 evidence rows, 0 tariff rows.

Which text-to-speech service gives the best Spanish ad voice, with commercial rights, for the least money?

  • First snapshot. Verdict: Insufficient evidence. 27 options, 0 evidence rows, 61 tariff rows.

Which text-to-speech service gives the best English narration of a 10-hour book for the least money?

  • First snapshot. Verdict: Insufficient evidence. 27 options, 0 evidence rows, 61 tariff rows.

Which text-to-speech service gives acceptable English audio at the lowest cost per million characters for a very large catalogue?

  • First snapshot. Verdict: Insufficient evidence. 25 options, 0 evidence rows, 55 tariff rows.

Which text-to-speech service speaks fastest and still sounds natural for a customer-service voice agent, for the least money?

  • First snapshot. Verdict: Insufficient evidence. 25 options, 0 evidence rows, 55 tariff rows.

Which text-to-speech service gives the best narration for a monetised English YouTube channel for the least money?

  • First snapshot. Verdict: Insufficient evidence. 27 options, 0 evidence rows, 61 tariff rows.

Data credits

Arena data: Arena (lmarena-ai), dataset lmarena-ai/leaderboard-dataset, https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset, licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Changes by Inadviser: the Arena scores (the Agent Arena IPS score and the text, text-to-image, text-to-video and search ratings) are rescaled to a 0-100 utility against fixed anchors set by Inadviser; the confidence interval and the observation or vote count are carried over; only a selection of models is shown on each board; where a board shows Arena's cost per task, it is shown unchanged. The dataset maintainers' method applies (https://arena.ai/blog/agent-arena-methodology/ for Agent Arena). Arena does not endorse Inadviser.

Energy cost per 100 km is computed by Inadviser from three inputs: consumption from the EEA, fuel prices from MITECO, and editorial electricity prices dated 2026-11-05 (EUR 0.12 and 0.20 per kWh at home, EUR 0.35 and 0.55 in public) that come from no dataset. The credits for the two datasets follow.