Creative intelligence.

Hierarchical Bradley-Terry leaderboard with partial pooling across the image studies. Higher Elo means stronger aggregate head-to-head performance; the whisker on each bar is the 95% confidence interval.

1. 1GPT Image 21095
2. 2Nano Banana Pro1094
3. 3Seedream V5 Pro1084
4. 4Meta Muse1070
5. 5Recraft V4.1 Pro1060
6. 6Ideogram v4994
7. 7FLUX.2 [pro]955
8. 8Midjourney V8.1940
9. 9Krea 2 Large921
10. 10Grok Imagine 1.0918

1. 

   ImageDataset

   Photoshop design trajectories294 annotated steps across 14 computer-use trajectories, recorded as five professional designers built advertising and fashion-editorial assets in Adobe Photoshop and the browser. Every step pairs a screenshot with a first-person thought, a structured action, and executable grounding: an MCP tool call, a keyboard shortcut, a menu path, or a coordinate click.294 steps · 14 trajectories · 5 designers · CC BY 4.0

   Hugging FaceAbout this dataset
2. 

   VideoDataset

   Video model annotations544 timestamped notes from professional video editors on 15 product videos made by Google Veo 3.1, Adobe Firefly Video, and Grok Imagine, all running the same five prompts. Every note carries a timestamp range, a comment, a quality dimension, and a severity rating.544 annotations · 15 videos · 3 models · CC BY 4.0

   Hugging FaceAbout this dataset

1. 
   07/10/2026ImageBattle

   Reve 2.1 trailed Seedream 5.0 Pro by 2 points on wins, then finished last on Elo.Four-model image battle across 10 briefs spanning portrait, environment, product, and lifestyle work, judged blind by working creative professionals.Read
   Tournament wins

   Seedream 5.0 Pro

   1st

   32.6

   Reve 2.1

   2nd

   30.2

   MAI Image 2.5

   3rd

   20.9

   ChatGPT Images 2.0

   4th

   20

   Share of tournaments won · Seedream 5.0 Pro leads Reve 2.1 by 2.4 pp
2. 
   07/10/2026Web DesignBattle

   Sol has taste. Fable takes direction.GPT 5.6 Sol, Claude Fable 5, Grok 4.5, and Muse Spark 1.1 on the same ten landing page briefs, judged blind by 9 working designers as live interactive pages.Read
   Win rate

   GPT 5.6 Sol

   1st

   63.3

   Claude Fable 5

   2nd

   49.3

   Grok 4.5

   3rd

   48.1

   Muse Spark 1.1

   4th

   39.3

   Share of 540 pairwise matchups won · 10 landing page briefs · July 2026. Split by brief structure the ranking inverts: Fable leads structured briefs at 1569 Elo to Sol's 1455.
3. 
   07/09/2026ImageBattle

   Seedream 5.0 Pro trails ChatGPT Images 2.0 by 8 points on wins, takes photorealism.Four-model image battle across 10 briefs spanning the capabilities ByteDance advertises. Seedream won photorealistic generation at 35.7% and finished first or second in 57% of its tournaments.Read
   Tournament wins

   ChatGPT Images 2.0

   1st

   35.9

   Nano Banana Pro

   2nd

   28.6

   Seedream 5.0 Pro

   3rd

   28

   Flux 2

   4th

   10.7

   ChatGPT Images 2.0 leads Seedream 5.0 Pro by 7.9 pp
4. 
   07/08/2026ImageBattle

   Meta Muse trailed ChatGPT Images 2.0 by 5 points on wins, then beat it on Elo.Four-model style-transfer battle across 10 briefs. Meta Muse took 29.5% of tournament wins and was the only model top-two on both wins and Elo.Read
   Tournament wins

   GPT Images 2.0

   1st

   34.1

   Meta Muse

   2nd

   29.5

   Nano Banana Pro

   3rd

   23.9

   Flux 2

   4th

   19.3

   Share of 88 tournaments won · ChatGPT Images 2.0 leads Meta Muse by 4.6 pp
5. 
   07/07/2026Web DesignBattle

   Write Fable a design spec and it wins 9 times out of 10Claude Fable 5 vs Claude Opus 4.8 on 5 real landing page and portfolio briefs, built in Claude Code, judged blind by 9 working designers.Read
   88.9%Best brief win rate

   11.1%Worst brief win rate

   51.1%Fable overall vs Opus

   Fable 5 vs Opus 4.8 · 90 blind matchups · July 2026
6. 
   06/29/2026VideoBattle

   Frontier AI video models are nearly tied. None of them nail physics yet.A blind head-to-head of Seedance 2.0, Grok Imagine, Veo 3.1, and Adobe Firefly Video, judged by 12 professional video editors across 10 prompts.Read
   Prompts won

   Seedance 2.0

   1st

   4

   Grok Imagine

   2nd

   3

   Veo 3.1

   3rd

   3

   Adobe Firefly Video

   4th

   0

   Prompts won out of 10 · 4 frontier video models · June 2026. Seedance 2.0 also led on average score, by +0.07 over Grok Imagine on a 5-point scale.
7. 
   06/17/2026ImageField Note

   Introducing Design Crit: we taught AI to judge design like a designer.Ten professional designers ranked four frontier image models across nine dimensions of real design work. The models can make the work. Nothing on the market could reliably judge it, until we trained on the right data.Read
   Agreement with panel

   Human designer

   Ceiling

   74.1%

   Trained on Design Crit

   Closes 46% of the gap

   61.1%

   Best off-the-shelf

   54.3%

   Chance

   50%

   Agreement with the five-designer majority · best off-the-shelf judge = HPSv2.1 · June 2026
8. 
   06/03/2026ImageBattle

   Ideogram v4 won 47.9% of typography matchups.10 designers, 4 models, 240 images. Spelling is solved. Typographic craft and client-readiness are where Ideogram v4 pulls away.Read
   Typography 1st-place rate

   Ideogram v4

   1st

   47.9%

   Gemini 3.1 Flash

   2nd

   30%

   FLUX.2 [max]

   3rd

   15.5%

   Grok Imagine 1.0

   4th

   15%

   Share of 1st-place finishes on typography prompts · 4 image models · 20 prompts × 10 reviewers · rounds 1 + 2
9. 
   06/01/2026ImageProfile

   Gemini reliably edits, but can it keep the rest of the image still?11 production-style sessions. Gemini made the edit 73% of the time, kept the rest of the image still 64%, held both in 55%.Read
   8 / 11Sessions that passed the edit-isolation test

   7 / 11Sessions that passed the pose-lock test

   6 / 11Sessions that passed both controllability tests

   Gemini controllability checks · 11 sessions · two tests per session (edit isolation + pose lock)
10. 
    05/29/2026Web DesignBattle

    Cursor took 60% of head-to-heads. Claude Code took 63% of client meetings.Four coding tools, 24 outputs, five working designers. The tool designers preferred to look at and the tool they'd put their name on turned out to be different.Read
    Client-ready

    Claude Code

    Most client-ready

    63%

    Antigravity

    53%

    Cursor

    Pairwise winner

    47%

    Codex

    Last

    30%

    “Would you present this to a client?” · 4 coding tools · 24 outputs · May 2026
11. 
    05/28/2026ImageProfile

    Gemini hit production-ready 24% of the time. One prompt pattern explains why.10 participants, 29 scored deliverables. The prompts that landed treated Gemini like a creative brief for a specific asset.Read
    7 / 29Deliverables Production Ready

    16 / 29Deliverables at Client V1+

    2 / 10Participants Production Ready on all 3 deliverables

    Gemini (Nano Banana Pro) production readiness · 10 participants, 29 scored deliverables
12. 
    05/27/2026ImageProfile

    Can Adobe Firefly edit like Photoshop?8 targeted edits across 4 designer sessions. Firefly cleanly resolved 1, drifted on 5, and missed 2.Read
    1 / 8Edits cleanly resolved

    5 / 8Edits landed partial

    2 / 8Edits unresolved

    Adobe Firefly Edit · 8 attempts across 4 sessions (Hero + Social per session)
13. 
    05/22/2026Cross-cuttingProfile

    The only prompt that got videos to production-ready in Adobe Firefly.4 designers, 3 deliverables each. The prompts that landed described physical direction, not aesthetic mood.Read
    1 / 4Videos production-ready, first pass

    2 / 4Social stills production-ready, first pass

    20Photoshop mentions across 4 sessions

    Adobe Firefly designer evaluation · 4 designers, 3 deliverables each
14. 
    05/21/2026Web DesignField Note

    In Claude Design, your opening prompt decides the ceiling.5 designers, 5 openings, 1 luxury brief. The first prompt set what each session could reach.Read
    Specificity score (0–1)

    First-prompt specificity by participant, Claude-scored · 5 designers, 1 brief
15. 
    05/20/2026Web DesignProfile

    Claude Design gets you 40%, Figma gets the rest.5 sessions, 5 designers, 1 real-world client brief. Strong as a starting structure, breaks under precision edits.Read
    60 → 100%Designers flagging layout & spacing, Edit 1 → Edits 4–5

    5 / 5Sessions where layout was the recurring failure mode

    ≤40%Designer verdict: use it to here, then hand off

    Claude Design designer evaluation · 5 designers, 1 real-world client brief
16. 
    05/18/2026ImageField Note

    With ChatGPT Images 2.0, "Text is solved." Typography isn't.42 sessions, 7 designers. ChatGPT Images 2.0 nails the typographic system, then breaks on the individual characters.Read
    +3 / +3 / +1Macro themes (hierarchy, brand fit, fonts) net positive

    −1 / −3Micro themes (legibility, size & weight) net negative

    0Designer mentions of size & weight as a strength

    Typography sentiment · 42 sessions, 7 designers, 6 briefs
17. 
    05/14/2026ImageProfile

    ChatGPT Images 2.0 won every head-to-head. Here's where it still breaks.41 sessions, 7 designers, 6 briefs. GPT Image 2 nails the concept, then breaks at production.Read
    5 / 41Sessions shipped from GPT alone

    33 → 59%Typography "no issues" climb

    60-65%Realism plateau, every iteration

    ChatGPT Images 2.0 production readiness
18. 
    05/13/2026ImageBattle

    The image-model leaderboard flips by brief.Four frontier image models, six brand campaigns, ranked blind by working creatives. GPT Image 2 wins the aggregate. Every other model owns a category.Read
    1st-place rate

    GPT Image 2

    1st

    40.7%

    Seedream 5.0 Lite

    2nd

    22.3%

    FLUX.2 [pro]

    3rd

    22.2%

    Gemini 3.1 Flash

    4th

    14.8%

    Share of 1st-place rankings across 6 brand campaigns · 4 frontier image models
19. 
    05/12/2026ImageBattle

    Krea 2 Large is the #2 style-transfer model, closing on GPT Image 2.Four-model style-transfer evaluation. Krea took #2 on style fidelity, 0.14 points behind GPT Image 2.Read
    Style Fidelity (avg)

    GPT Image 2

    1st

    3.53

    Krea 2 Large

    2nd

    3.39

    Gemini 3 Pro

    3rd

    2.74

    Seedream 5.0 Lite

    4th

    2.42

    Style Fidelity average rating · Krea 2 Large takes #2, 0.14 points behind GPT Image 2
20. 
    05/06/2026ImageBattle

    Seedream 5.0 Lite swept the field on product detail shots.A blind head-to-head against the leading image models from Google, OpenAI, and Black Forest Labs, evaluated by professional creatives.Read
    Win rate

    Seedream 5.0 Lite

    1st

    63.9%

    Gemini 3 Pro

    2nd

    52.8%

    GPT Image 1.5

    3rd

    44.4%

    FLUX.2 [max]

    4th

    38.9%

    Pairwise win rate · 4 leading image models · March 2026
21. 
    05/05/2026Cross-cuttingField Note

    Creatives keep telling us the same thing about AI: every output looks the same.12 models, 5 creative domains. One repeated complaint from working evaluators: the work all looks the same.Read
    Convergence (best-practice) and divergence (steerability) as orthogonal signals.
22. 
    04/28/2026VideoProfile

    Grok Imagine is the "Polisher" model. Hand off the early rounds, bring it in for refinement.The biggest phase-over-phase climb of any video model in the study. 3rd at ideation, 1st at refinement.Read
    Win rate

    Ideation

    3rd place

    46%

    Mockup

    44%

    Refinement

    1st place

    56%

    Grok Imagine win rate · +10pp climb from ideation to refinement
23. 
    04/23/2026Cross-cuttingField Note

    The creative process has 3 phases. AI performs very differently in each.Ideation, mockup, refinement. AI fits differently at each phase, and the best creatives know where to hand off.Read
    Ideation

    Loose grip

    Mockup

    Narrowed

    Refinement

    Firm grip

    How tightly creatives hold control across phases · Qualitative
24. 
    04/22/2026VideoProfile

    Veo 3.1 is the "Creative Director" model. Use it early, but hand off before refinement.61% win rate at ideation. 39% at refinement. The clearest model profile in our video evaluation.Read
    Win rate

    Ideation

    Peak

    61.1%

    Mockup

    55.6%

    Refinement

    Last place

    38.9%

    Veo 3.1 win rate · −22.2pp drop from ideation to refinement
25. 
    04/21/2026Cross-cuttingField Note

    Solo creatives are earning more with AI and staying independent.Higher earning potential, more projects, no new hires. The survey from working independents.Read
    of independent creatives report higher earning potential since adopting AI

    26% no · 8% other

    Survey · Independent creatives on Contra
26. 
    04/14/2026Web DesignBattle

    We tested 4 AI models with professional web designers. Claude won, but not the way you'd expect.Claude Opus 4.6, Gemini 3.1 Pro, ChatGPT 5.3 Codex, Qwen 3.5. The winner shifted at every phase.Read
    Leading win rate

    Ideation

    Claude leads

    79.8%

    Mockup

    Gemini takes over

    68.9%

    Refinement

    Claude narrows gap

    60%

    Per-phase leader shifts · Preview of the Human Creativity Benchmark
27. 
    04/08/2026Cross-cuttingField Note

    AI isn't replacing creative professionals. It's making the best ones better.Survey of high-earning independent creatives. What they actually do with AI on real client work.Read
    of AI output makes it to final deliverables

    The rest is stripped, reworked, or scrapped

    Dominant survey response · Independent creatives