←
Rapidata210 posts
Rapidata avatar

Rapidata

@RapidataAI

6K+ human judgements/minute via API for Online RLHF and Benchmarks. Advancing SOTA GenAI research through weekly, open benchmarks → benchmark.ai

◷ Joined June 2024▧ 113 Photos and videos
Rapidata avatar
Rapidata@RapidataAI·Nov 10

Visual realism ≠ spatial controllability. Generating convincing scenes is one thing, and being able to navigate them precisely is another. This matters for most use cases of world models. @theworldlabs Atlas shows impressive capabilities here, ranking #1 across 600 prompts and 513K+ human judgments.

Great job @theworldlabs teams! 🚀 Camera control is so critical for many creative and design work. I’m very excited by Atlas’ strong results!

Rapidata avatar
Rapidata@RapidataAI·14h

OpenAI has held the image generation SOTA for over 2 years. Now @MicrosoftAI’s MAI Image 2.7 takes #3 on Benchmark.ai. It's the first non-OpenAI model to break into the top 3. 👀 This benchmark contains 7.5M+ human judgements, over 4 dimensions, and 1.5K prompts. All data is transparent.

0:33
Rapidata avatar
Rapidata@RapidataAI·14h

GPT-6.2 Astra enters our LLM-to-3D leaderboard at #2 out of 32, making it @OpenAI’s highest-ranked model 🥈 Each model writes three.js code from a text prompt. We ranked their 3D outputs using 1.89M+ human judgements across 500 prompts. Results, human votes, and match-ups available on Benchmark AI

0:41
Rapidata avatar
Rapidata@RapidataAI·Nov 16

Seedream 5.5 enters at #2 for character preservation 👀 We apply five successive edits to an image, then humans judge which model best preserves the person’s identity. 200 prompt chains and 381K+ votes across models. Some results below 👇

0:57
Rapidata avatar
Rapidata@RapidataAI·Nov 13

With @thinkymachines Inkling 2 small now available they are the #2 open weight model in 3D generation and #1 western open weight model! Kimi is roughly 5x bigger making Inkling a much more reasonable choice to run on prosumer hardware.

0:41
Rapidata avatar
Rapidata@RapidataAI·Nov 13

Meta’s Muse Image flattens the Pareto frontier for character preservation: #3 at $0.01/edit. We apply 5 successive edits to a photo, then asked 300k+ humans which model best preserves the person’s identity. Results on benchmark.ai 👇

1:00
Rapidata avatar
Rapidata@RapidataAI·Nov 12

🚨 Gemini 4 Hydrogen enters at #4 and Pareto frontier on our LLM-to-3D benchmark. That puts 3 Gemini models in the top 4, across 25 models and 1.48M+ human judgements. See what it builds 👇

0:46
Rapidata avatar
Rapidata@RapidataAI·Nov 12

Meta’s smallest model is surprisingly good at 3D 👀 @AIatMeta’s Muse Ember lands on the Pareto frontier of our LLM-to-3D benchmark, combining strong performance with the lowest price on the frontier. 1.38M+ human votes across the benchmark. Some match-ups below. All votes and outputs openly available on Benchmark AI.

0:47
Rapidata avatar
Rapidata@RapidataAI·Nov 10

🔥 Gemini 4 Carbon was released a few hours ago and ranks #1 in our LLM-to-3D Benchmark. This benchmark evaluates how good LLMs are at constructing 3D objects; it includes 1.3M+ human judgments. As usual, you can find all data available on Benchmark AI.

0:54
Rapidata avatar
Rapidata@RapidataAI·Nov 10

After 5 edits, does the person still look the same? Ideogram 4.5 ranks #1 among 20 models for preserving facial identity! 🥇 We benchmarked character identity drift across 199 photos, applying 5 consecutive edits and evaluating only the resulting faces. This benchmark holds 322K+ human judgments. All outputs, judgments and methodology are publicly available on Benchmark AI.

0:55
Rapidata avatar
Rapidata@RapidataAI·Nov 10

Fable 5.2 enters our SVG benchmark at the #1 position. The benchmark contains 4M human judgments across prompt alignment, coherence, and preference. All outputs, votes, and methodology are public on benchmark.ai/svg.

0:41
Fei-Fei Li avatar
⟳ Rapidata retweeted
Fei-Fei Li@drfeifei·Nov 9

Great job @theworldlabs teams! 🚀 Camera control is so critical for many creative and design work. I’m very excited by Atlas’ strong results!

Atlas ranks #1 in @RapidataAI's camera-movement benchmark, which evaluates how well models follow camera instructions. The evaluation compares 19 world and video models across 600 prompts, with 513K+ human judgments. More on the results and methodology below.

Post attachment
Rapidata avatar
Rapidata@RapidataAI·Nov 9

World Labs’ Atlas takes #1 in camera movement! 🥇 We evaluated 19 world and video models across 600 prompts, collecting 513K+ human judgments to assess how accurately each model follows camera movement instructions. All outputs, matchups, judgments, and methodology are public on Benchmark AI.

0:41