logo
Contact Us
Contact Us

Blogs

Covering everything you need to know in order
to build your AI breakthrough

Hazel Gao
23 Dec, 2025/Research

EditReward: Abaka AI’s Human-Aligned Reward Model for Image Editing

Discover EditReward — Abaka AI’s human-aligned reward model that outperforms GPT-5 on image-editing evaluation. Powered by expert-annotated data and multidimensional scoring.

Nadya Widjaja
26 Dec, 2025/General

Annotate a Video Poorly and No Amount of Data Will Save Your Mode

Most video models fail not because of data size, but because of coarse labels. Learn how label granularity directly impacts accuracy, generalization, and multimodal reasoning.

Abaka
15 Aug, 2025/Weekly Insights

Abaka Pulse: AI Agents & Coding Data Insights | Aug 1-Aug 12

Explore AI agent advancements: datasets, evaluation, CGAT tool, VeriGUI paper, ACL 2025 recap, and Bay Area hiring.

Abaka
30 Nov, 2023/Why Choose Abaka

Abaka AI vs Surge AI: Comprehensive AI-Powered Data Labeling and Content Moderation Platform

Explore Surge AI, the comprehensive AI-powered data labeling and content moderation platform. Learn about its features, pricing, and user feedback for managing and annotating large datasets.

Hazel Gao
23 Dec, 2025/Research

EditReward: Abaka AI’s Human-Aligned Reward Model for Image Editing

Discover EditReward — Abaka AI’s human-aligned reward model that outperforms GPT-5 on image-editing evaluation. Powered by expert-annotated data and multidimensional scoring.

Hazel Gao
23 Dec, 2025/Research

Transforming Research Papers into Frontier-Level Reasoning Benchmarks

Learn how frontier-level reasoning benchmarks are built by transforming real research papers into self-contained, multi-step reasoning tasks. Designed to resist shortcuts from GPT-5.1, Gemini 3 Pro, and Claude 4.5, this pipeline sets a new standard for evaluating true reasoning ability.

Hazel Gao
16 Dec, 2025/Research

Breaking the Code Intelligence Barrier: How SWE-Bench Pro and Abaka AI Enable Real-World Coding Intelligence

SWE-Bench Pro marks a new era in evaluating coding intelligence. Learn how Abaka AI builds high-fidelity, reproducible, human-validated coding datasets that mirror real engineering complexity — and why they are essential for training and benchmarking next-generation code-reasoning models.

Johnson Jiang
26 Sept, 2025/Research

Abaka AI’s VeriGUI: Building Trustworthy Agent Data

Discover how Abaka AI‘s VeriGUI project is setting the standard for high-quality agent data. Learn about our unique, dual-engine pipeline that combines real-world task construction with a multi-layered, adversarial QA process to produce data that is both challenging and trustworthy for advanced AI agents.

Hazel Gao
23 Sept, 2025/Research

Decoding "Nano Banana": Key to Next-Gen Image Editing—Fine-Grained Instruction Data

Next-gen AI image editing succeeds with fine-grained data—Nano Banana proves this. Abaka AI’s datasets turn simple prompts into vivid scenes, enhancing Part Editing, Object Replacement and style control.

Abaka
25 Aug, 2025/Research

Abaka 1B+ High-Quality Question Bank: Fuel for AI Revolution

Abaka AI’s 1B+ high-quality question bank solves AI training data bottlenecks. Features 3-tier verification, covers K12 to university, multiple languages, and pairs with academic databases to boost model performance.

Abaka
25 Aug, 2025/Research

Abaka AI COIG-Writer: Adding Human Touch to Chinese AI Writing

AI-generated Chinese content lacks depth & emotion. Abaka AI's COIG-Writer dataset uses a Query-Thought-Answer structure to enhance logic, narrative & personality. Open-sourced on Hugging Face. Custom solutions available.

Abaka
25 Aug, 2025/Research

Talking-Head Video Data: Core for Multimodal AI’s Speaking Skills

Talking-head video data (synced audio-visual) powers multimodal AI’s speaking skills. Abaka AI solves public dataset gaps with custom, high-quality datasets for digital humans, TTS, and AIGC.

Page 1 of 1

Never miss an update

Get all the news and updates about Abaka, directly to your inbox.