AI Model Power Rankings

Still top of the Artificial Analysis index at 58 and #1 on Arena WebDev. Two new rivals arrived this week, and neither beat it.

Finally on sale, and still #1 with Arena text voters. Book it at $4 / $20 in your budget, though, because that's the price from December 1.

Scores 56 on the Artificial Analysis index, ahead of Argon's 53 at the same $2 / $10. Pushed to #3 only by Argon's text-board lead.

A quiet week. Still the default in Codex and the best value in OpenAI's lineup.

Meta's first frontier model debuts at #2 on Arena text. It costs as much as Opus 5.5 and scores six points lower on the Artificial Analysis index, so it isn't the better buy.

Still superb on the hardest problems, but at 2.5x Opus 5.5's price it's a specialist pick now.

Argon now matches it on the Artificial Analysis index for a fifth of the list price. Keep it for long automation jobs, where it still leads.

Its second week keeps it as the best of the $0.10 models. Nothing new arrived to challenge it.

Still a text top-10 model at Flash prices, but the January price doubling is getting closer.

Still the strongest model from outside the US labs, and still ahead of Meta's new open model on the Artificial Analysis index.

Muse Spark Open arrived to challenge it and fell short on general smarts and coding. Still the open-weight model to beat.

Meta's own Muse Wonder now takes the spotlight. It's still decent value, and it actually edges Wonder on WebDev.

Still #1 on Arena WebDev. Meta's own launch table has it well ahead of Muse Wonder on ProgramBench, 70.5 to 50.0.

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

The default in Codex. Most of Astra's coding for a fifth of the price.

Still #2 on WebDev, but Sol does most of the same work for much less.

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

Now you can test it yourself. Still outside the WebDev top five, and Bloomberg reports it does worse on some practical coding jobs than its benchmarks suggest.

#15 on WebDev and #12 in Arena's agent rankings. A solid debut, but not one to switch your coding agent for at $4 / $20.

Still a popular pick for cheap coding agents.

The open-weight coding pick. On Meta's own table it beats Muse Spark Open on Terminal-Bench.

Independent Terminal-Bench runs land in the mid-30s. That's plenty for cheap sub-agents.

$0.10 / $0.50 under 100K tokens and the strongest independent results at that price. Mind the 5x jump above 100K.

Top-10 text quality at Flash prices until January. Then it doubles.

Astra-level results at $2 / $10, with Batch at half that. Its score drops in December, when the price doubles.

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

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

Same $0.10 / $0.50 as Haiku 5.5, with weaker independent results. Worth a look if you're already on OpenAI.

Strong conversation at $1.25 / $4.25.

$0.15 / $0.47 with good multilingual skills. Qwen 4 is still undated.

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

Meta's own launch table puts GLM-5.3 ahead of Muse Spark Open on the Artificial Analysis index, 45 to 39. Still #1 here.

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

The strongest US open-weight model by Meta's count, and it leads Meta's tool-use tests. It trails the Chinese leaders on general smarts and coding, and it runs under Meta's own license.

Apache 2.0 and small enough for real-world hardware.

Still leads the text-to-image board. Nothing this week came close.

Microsoft's in-house model stays just ahead of Google's newcomer.

A week in, it's still top five at about half Nano Banana 2's price. The best pick for editing in chat.

Still the indie favourite for aesthetics.

Still top-tier on its low setting.

Powers the CapCut/Dreamina creator ecosystem.

A quiet week for video. Google still holds the top two text-to-video spots.

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.

Near the top of Artificial Analysis's video board, lower with Arena voters. Still the tinkerer's pick.

Professionals still choose it for camera control.