AI Model Power Rankings

Still 58 on the Artificial Analysis index and #1 on Arena WebDev. Anthropic chose it to run Claude Duet, its new always-on agent.

Still #1 on Arena text, with a preliminary 1524. A quiet first full week on sale. Budget for $4 / $20 from December 1.

Still 56 on the Artificial Analysis index, three points clear of Argon at the same $2 / $10.

No change. Still the best value in OpenAI's lineup and the default in Codex.

Holds #2 on Arena text in its second week, but 52 on the Artificial Analysis index is still six behind Opus 5.5 at the same price.

Nothing new. Still a specialist pick for the hardest problems at 2.5x Opus 5.5's price.

GPT-6 Pro in ChatGPT now splits a task across parallel agents. That's useful, but it's Pro-only and OpenAI published no numbers to judge it by.

New $0.15 models arrived this week. At $0.10 / $0.50 it's still the cheapest of the strong small models.

Still good, but the new Chinese Flash models now cost a fifth as much per independent index task, and its January price rise is coming.

New entry. The 1017 snapshot ties GLM-5.3 at 45 on the Artificial Analysis index, at a lower hosted price and under an MIT license.

Now edged by DeepSeek's cheaper Flash on the Artificial Analysis index, 45 to 44. Still a favourite for agent work.

Now shares the open-weight top score of 45 with DeepSeek V4.1 Flash, and still has the deeper Arena vote record.

Still #1 on Arena WebDev. Nothing new this week threatened it.

Top four on WebDev at half Opus 5.5's price. Still the pick for well-scoped fixes.

OpenAI says Codex sub-agents can now reuse context and keep it cached for longer, which should cut repeat costs on long jobs.

Still #2 on WebDev. Sol does most of the same work for a fifth of the price.

Excellent code, but two cheaper Anthropic models do most jobs as well.

Still outside the WebDev top five. Early reviews rate its writing above its coding.

A quiet second week. Still mid-table on WebDev at Opus 5.5's price.

Still a popular cheap coding agent. DeepSeek and MiniMax have both entered WebDev, so check back next week to see where they land.

Still the open-weight coding pick until the new WebDev entries have enough votes.

Independent Terminal-Bench runs in the mid-30s. Plenty for cheap sub-agents.

$0.10 / $0.50 under 100K tokens and 43 on the Artificial Analysis index at max effort. Still #1, though the gap is now small.

New. A 45 on the index for about $0.26 per task. It's half price off-peak, but $0.30 / $1.20 in peak hours.

Near the top of the index at $2 / $10 until December, with Batch at half that.

Near-Astra coding at $2 / $10, with cached input at $0.10.

Down three places. About $1.24 per Artificial Analysis index task is now hard to defend, and the price doubles in January.

The highest index score at $2 / $10, but it uses lots of tokens, which eats into the savings.

New. The cheapest per index task this week at about $0.22, but with a 93% hallucination rate, so don't use it where you can't check every answer.

Same price as Haiku 5.5 with weaker independent results, and now cheaper rivals from China too.

Strong conversation at $1.25 / $4.25.

$0.15 / $0.47 with good multilingual skills, but DeepSeek and MiniMax now charge about the same.

Very fast diffusion text generation at $0.20 / $0.75.

Now tied with DeepSeek at 45 on the Artificial Analysis index. It keeps #1 on its Arena record and its cheap Flash sibling.

New. The October 17 snapshot matches GLM-5.3 on the index and ships under MIT, a cleaner license than GLM's own.

Level with GLM-5.3 on Arena, on fewer votes. Another quiet week.

Still the strongest US open-weight model at 39 on the index. Reflection's Beam arrived behind it.

New. Apache 2.0 weights and published training recipes, but 30 on the index puts it well behind the leaders. There's no hosted API yet.

Still the one on this list that runs on hardware you might own. See our guide to running AI on your own computer.

No new text-to-image entries on Arena this week. Still the clear leader.

Still just ahead of Google's newest.

Still top five at about half Nano Banana 2's price.

Still the indie favourite for aesthetics.

Still top-tier on its low setting.

Powers the CapCut/Dreamina creator ecosystem.

Another quiet week for video. Google holds the top of the board.

The safest all-rounder, with native audio and 4K.

Holds #3, still on a small vote count.

The value king at about 11–14 cents per second, plus storyboard mode.

Top five in agent mode. Fast and fun.

Still the best realistic human motion for creators.

Still the tinkerer's pick for custom pipelines.

Professionals still choose it for camera control.