AI Model Power Rankings

Now #1 with the voters as well as the benchmarks, on LMArena's text and WebDev boards. The text lead is provisional on a small sample, but nothing has knocked it off yet.

Still in the text top five and #3 on WebDev. Brilliant, but hard to justify at 2.5x Opus 5.5's price.

Lost WebDev #1 to Opus 5.5 and trails on the text board. It still leads on automation work, which is where it earns its price.

Voters rank it alongside Claude Opus 5 for web coding at a fraction of the cost. The best-value model near the frontier.

Holding a text top-10 spot for 75 cents per million input tokens. The price doubles on January 1.

The model is fine. Meta's Muse agent being locked out of Amazon is a product problem, not a quality one.

Still in preview and outranked by its Flash sibling. Google says Gemini 4 is in post-training, which is the upgrade this slot needs.

Top 10 on WebDev and level with Gemini 3.1 Pro on the text board. Still the strongest model from outside the US frontier labs.

Its first WebDev showing is respectable, just outside the top 12. Good value, but not in the lead pack.

Still the best-proven open-weight model, but Xiaomi's MiMo now matches it on the text board.

Early testers report it keeping pace with Sol on some coding evals at a fraction of the price. We'd like independent numbers before we move it up.

Cheap and multilingual. Alibaba's voice price cuts this week make its wider stack even better value.

#1 on Arena WebDev, well clear of GPT-6 Astra, and top of Terminal-Bench 4.0. The default pick for coding.

Now #2 on WebDev and still the strongest when the job involves driving a terminal.

#3 on WebDev. Excellent, but Opus 5.5 is better and cheaper for most code.

Enters the WebDev top five on real votes, on par with Claude Opus 5. Live in Codex and GitHub Copilot.

Top 10 on WebDev with a big vote count behind it. A popular choice for cheap coding agents.

The open-weight coding pick, though MiMo-V2.6-Pro scores level with it on WebDev.

Great for quick edits and autocomplete-style work. Wait for Gemini 4 if you need an agent.

Edges out 4.6 on WebDev, but 38% on Terminal-Bench 4.0 still rules it out for serious agents.

Top-10 text quality at Flash prices, so enjoy it before January.

Arena voters now back up the pitch: Opus 5-level web coding for a fifth of Astra's price.

$0.10 / $0.50. Early coding results look strong, but they're mostly OpenAI's and testers' numbers so far.

Strong conversation at $1.25 / $4.25, and the Contributor tier is even cheaper.

$0.15 / $0.47 with multilingual strength, and Alibaba's audio APIs just got much cheaper too.

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

Its $6 output tokens are among the cheapest at this quality level.

Still the strongest open-weight model on the most evidence. It has a custom license, so read it.

Its first Arena scores put it level with GLM-5.3 on both text and WebDev. Fewer votes so far, but a serious debut.

Apache 2.0 and small enough for real-world hardware.

Still holds the top three spots on the leaderboard, with the gap unchanged.

Microsoft's in-house model remains the best non-OpenAI option.

Moves up to fifth on the leaderboard. The indie favourite for aesthetics.

Slipped a place as more votes came in, but still top six on its low setting.

Ranks lower on raw generation, but it's the best for editing an image through conversation.

Powers the CapCut/Dreamina creator ecosystem, and Alibaba's Qwen-Image 3.0 Pro is now level with it.

Google holds the top two text-to-video spots with Omni Flash and its 1.1 update.

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

Holds #3 on the leaderboard, though on a small vote count.

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

Up to #4 in agent mode. Fast and fun.

Dipped below its own predecessor as votes piled in. Still the best realistic human motion for creators.

Dropped out of the top five as its vote count doubled. Still the tinkerer's pick.

Professionals still choose it for camera control.