Week 39, 2026Published September 25, 2026

AI Model Power Rankings

Opus 5.5 takes the crown in the busiest launch week of the year. Tuesday, September 22 brought five launches in one day: Anthropic's Claude Opus 5.5, OpenAI's GPT-6 Sol and Luna, and Xiaomi's MiMo-V2.6 Pro and Flash. That's enough to reshuffle the whole board. This is our debut edition, so there are no movement arrows yet. From next week, every change gets tracked.

RiserGemini 3.8 Flash

A 'fast' model sitting in the LMArena top 10, above Google's own Pro model. Grab the intro price before it doubles on January 1.

FallerGemini 3.1 Pro

Still in preview, and now beaten by its cheaper Flash sibling. Google needs a new Pro model.

01Power Rankings

The best general-purpose AI models right now, with quality, price and momentum rolled into one score.

#Movement±ModelPrice / 1M tokensScore
1 · Claude Opus 5.5Anthropic

Top of the Artificial Analysis Intelligence Index and the best coder on the board, for less than Opus 5 cost. A new champion on day one.

$4 in / $20 out 97
2 · Claude Fable 5.1Anthropic

Still the heavyweight, leading the Arena agent leaderboard. You pay 2.5x Opus 5.5's price for the last few percent.

$10 in / $50 out 95
3 · GPT-6 AstraOpenAI

OpenAI's best, and still #1 on automation benchmarks. It lost the index lead to Opus 5.5 this week.

$10 in / $50 out 94
4 · GPT-6 SolOpenAI

The value story of the week: Astra-style training at $2/$10, with roughly half as many mistakes as GPT-5.6 Sol.

$2 in / $10 out 92
5 · Gemini 3.8 FlashGoogle

Arena top 10 for 75 cents per million input tokens, which is absurd value until January.

$0.75 in / $3.75 out 90
6 · Muse Spark 1.3Meta

Meta's quiet contender. Voters love how it talks, and the price didn't budge.

$1.25 in / $4.25 out 89
7 · Gemini 3.1 ProGoogle

Still good, but lapped by its own Flash sibling. We want to see a Gemini 4 Pro.

$2 in / $12 out 87
8 · Kimi K3Moonshot AI

The highest-ranked model outside the US frontier labs. It deserves more attention than it gets.

Not published 86
9 · Grok 4.7xAI

A steady update, and cheap output tokens help. Agentic coding is still well behind the leaders.

$2 in / $6 out 84
10 · GLM-5.3Z.ai · open

The best model you can download and run yourself, if you have the hardware for 753B parameters.

Not published 84
11 · GPT-6 LunaOpenAI

Not a genius, but at $0.10 per million input tokens it'll power half the chatbots on the internet by Christmas.

$0.1 in / $0.5 out 82
12 · Qwen3.8-FlashAlibaba

Cheap, multilingual and now multimodal through Omni-Flash. A sensible budget pick.

$0.15 in / $0.47 out 78

02Coding

For developers and 'vibe coders': writing, fixing and shipping real software.

#Movement±ModelPrice / 1M tokensScore
1 · Claude Opus 5.5Anthropic

66.4% on Terminal-Bench 4.0 and #1 on Arena WebDev. It's the default pick for coding.

$4 in / $20 out 98
2 · GPT-6 AstraOpenAI

57.9% on the same benchmark, and strongest when the job involves driving a terminal.

$10 in / $50 out 94
3 · Claude Fable 5.1Anthropic

55.8% there. It's brilliant, but Opus 5.5 is both better and cheaper for most code.

$10 in / $50 out 93
4 · GPT-6 SolOpenAI

A huge jump over GPT-5.6 Sol, and it's live in Codex and GitHub Copilot.

$2 in / $10 out 90
5 · Kimi K3Moonshot AI

A popular choice for cheap coding agents.

Not published 84
6 · GLM-5.3Z.ai · open

The open-weight coding pick.

Not published 83
7 · Gemini 3.8 FlashGoogle

Great for quick edits and autocomplete-style work.

$0.75 in / $3.75 out 82
8 · Grok 4.7xAI

38% on Terminal-Bench 4.0. Fine for snippets, not for agents.

$2 in / $6 out 74

03Best Value

The most capability per dollar. If you're building an app or watching a budget, start here.

#Movement±ModelPrice / 1M tokensScore
1 · Gemini 3.8 FlashGoogle

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

$0.75 in / $3.75 out 95
2 · GPT-6 LunaOpenAI

$0.10 / $0.50, and the GPT-6 lineage behind it shows.

$0.1 in / $0.5 out 93
3 · GPT-6 SolOpenAI

Near-frontier quality for a fifth of Astra's price.

$2 in / $10 out 91
4 · Muse Spark 1.3Meta

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

$1.25 in / $4.25 out 88
5 · Qwen3.8-FlashAlibaba

$0.15 / $0.47 with multilingual strength.

$0.15 in / $0.47 out 86
6 · Mercury 2.5Inception

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

$0.2 in / $0.75 out 80
7 · Grok 4.7xAI

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

$2 in / $6 out 78

04Open-Weight Watch

Models you can download, run privately and customise.

#Movement±ModelPrice / 1M tokensScore
1 · GLM-5.3Z.ai · open

The strongest open-weight model right now, with a custom license, so read it.

Not published 92
2 · MiMo-V2.6-ProXiaomi · open

Xiaomi keeps shipping. It launched this week, so treat it as provisional.

Not published 85
3 · Ternary Bonsai 2 27BPrismML · open

Apache 2.0 and small enough for real-world hardware.

Not published 78

05Image Generation

Making pictures from text, based on LMArena text-to-image votes plus our hands-on notes.

#Movement±ModelPrice / 1M tokensScore
1 · GPT Image 2.5OpenAI

It holds the top three spots on the leaderboard. Not a contest right now.

— 97
2 · MAI-Image 2.6Microsoft AI

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

— 88
3 · Reve 2.1Reve

The indie favourite for aesthetics.

— 85
4 · Grok Imagine Image 2.0xAI

Top six even on its low setting.

— 84
5 · Gemini 3.1 Flash ImageGoogle

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

— 82
6 · Seedream 5.0 ProByteDance

Powers the CapCut/Dreamina creator ecosystem.

— 80

06Video Generation

Text-to-video, and what to use now that Sora is gone.

#Movement±ModelPrice / 1M tokensScore
1 · Gemini Omni 1.1 FlashGoogle

#1 on LMArena text-to-video at 1516, though with only a few votes so far.

— 95
2 · Veo 3.1Google

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

— 92
3 · FLUX 3 VideoBlack Forest Labs

Black Forest Labs arrives in video at #3.

— 90
4 · Kling 3.0Kuaishou

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

— 89
5 · Seedance 2.5ByteDance

The best realistic human motion for creators.

— 88
6 · Grok Imagine Video 1.5xAI

Fast and fun, ranking #4 in agent mode.

— 86
7 · Wan 3.0Alibaba

The tinkerer's pick.

— 84
8 · Runway Gen-4.5Runway

Professionals still choose it for camera control.

— 82

The Wire · what happened this week

  1. OpenAI's Sora API shuts down today. The Sora app already closed on April 26. Anyone still building video on Sora needs to move.

  2. Anthropic releases Claude Opus 5.5 at $4/$20 per million tokens, 20% cheaper than Opus 5. Sonnet 5.5 and Haiku 5.5 are due 'in the weeks ahead'.

  3. OpenAI launches GPT-6 Sol ($2/$10) and GPT-6 Luna ($0.10/$0.50) across ChatGPT, Codex and the API.

  4. Xiaomi open-sources MiMo-V2.6-Pro and MiMo-V2.6-Flash.

  5. xAI ships Grok 4.7 at the same $2/$6 price as 4.6.

  6. Alibaba adds Qwen3.8-Omni-Flash, a multimodal model priced at $0.15/$0.47.

  7. PrismML releases Ternary Bonsai 2 27B, a compressed model with open weights under the Apache 2.0 license.

  8. Moonshot previews Kimi K2.8, and Sakana AI launches Fugu Max.

Scores are our editorial blend of public leaderboards (LMArena, Artificial Analysis and published benchmarks), price and hands-on use. Read the methodology →