Innodata

09/30/2026

Innodata has opened a new motion-capture AI lab designed to generate the high-quality data needed to train the next generation of humanoid robots and physical AI systems.

Developed in collaboration with Vicon, the lab uses high-precision optical motion capture technology to measure movement with sub-millimeter accuracy, capturing 3D motion directly from humans and robots.

The lab expands Innodata’s end-to-end physical AI capabilities, supporting:
• High-quality 3D motion capture data for AI training
• Custom and off-the-shelf motion datasets
• Robot performance evaluation and benchmarking
• Independent validation of internal robot telemetry
• Data collection through model evaluation and safety assurance

“Physical AI is growing faster than any other segment in AI, but every robotics team hits the same wall: there isn't enough real-world interaction data, and what exists is expensive and slow to produce,” said Rahul Singhal, CEO of Innodata.

Read the full announcement: https://investor.innodata.com/news/news-details/2026/Innodata-Opens-Motion-Capture-AI-Lab-to-Help-Humanoids-Move-Like-Humans/default.aspx

07/15/2026

An AI agent can perform well in a demo and still fail inside an enterprise workflow.

The difference often comes down to workflow fit:

Does the agent have the right context? Clear decision boundaries? Access to the right systems? Defined controls and escalation paths? A measurable standard for success?

Our latest article outlines six questions enterprises should answer before moving an AI agent into production.

Read here:

Explore six questions for aligning agents with enterprise context, controls, systems, evaluation, and oversight.

07/14/2026

Most AI benchmarks test models on short, isolated tasks.

But that is not how AI systems are used in practice.

Interactions are long, messy, and multimodal. Users revise instructions. Important details are buried in documents, images, or prior turns. Context carries across the full conversation.

Innodata’s Long Context & Complex Interaction (LCCI) Benchmark is designed to evaluate how AI models perform under those conditions.

In our latest article, we share preliminary findings from LCCI and explore why long-context LLM evaluation needs to look beyond aggregate scores and toward the specific failure modes that matter in real-world model performance.

The article also links to our public preview dashboard, where readers can explore the benchmark results interactively.

Read more:

Explore Innodata’s LCCI Benchmark for long-context LLM evaluation across multi-turn conversations, multimodal inputs, and real-world model failures.

06/18/2026

Innodata is pleased to announce the appointment of Jayant Chauhan as Executive Vice President and Chief Financial Officer, effective July 6, 2026.

Jayant brings more than two decades of experience building and leading finance functions for fast-growing global technology companies.

Read the full announcement:

Jayant Chauhan Brings More Than Two Decades of Experience in Finance and Scaling High-Growth Technology Companies

05/18/2026

The Innodata GenAI Summit is happening this week in London 🏙️

On May 21, we’ll bring together AI builders, researchers, executives, and technical leaders for a focused day on the systems, infrastructure, and deployment challenges shaping AI today.

The conversations will span world models, autonomous systems, physical AI, evaluation, safety, infrastructure, and what it takes to bring advanced AI systems into production.

We're looking forward to an incredible day in London with leaders from across the AI ecosystem.

📍 London
📅 May 21, 2026
🔗 Register: https://world.aiacceleratorinstitute.com/location/innodatasummit/register

05/13/2026

LLMs may know cultural facts, but can they apply cultural understanding when context is implied? 🌎

We’re sharing findings from the Innodata Cultural Alignment Benchmark (ICAB), a multimodal evaluation built to assess how well LLMs understand culture beyond surface-level knowledge.

ICAB examines whether models can navigate implicit cultural context across languages, locales, and real-world scenarios, from workplace norms to family dynamics to culturally specific visual prompts.

The results point to a clear gap: models perform better when cultural expectations are explicit, but struggle when they need to infer nuance from context.

Read the blog and explore the interactive results dashboard here: https://innodata.com/cultural-alignment-llms-benchmark-applied-understanding/

Innodata’s ICAB benchmark evaluates how well LLMs understand implicit cultural context across languages, locales, and multimodal tasks.

04/28/2026

Exciting news for Innodata Federal 🇺🇸

We’re thrilled to announce that Innodata Federal has been accepted into the CDAO Tradewinds platform as an awardable entity- meaning we’re now eligible to receive and deliver AI-driven projects directly through this prestigious federal marketplace.

This milestone marks a significant step forward in expanding how we support government agencies with mission-critical AI solutions- faster, more efficiently, and at scale.

Explore what we’re building in federal AI: https://innodata.com/federal/

04/17/2026

AI doesn’t always pick the best answer.
In fact, it often settles for “good enough.”

Why? Because in large systems, the sheer volume of average options can overwhelm truly high-quality ones.

In our latest article, Innodata’s Senior AI Researcher Esther Derman unpacks this hidden problem and explains how new approaches in reinforcement learning are helping rebalance exploration and precision.

Drawing on her research presented at ICLR, she shows how AI systems can move beyond proxy-driven results toward real-world performance.

Read more:

Why AI systems favor average results over the best ones, and how robust reinforcement learning improves real-world performance.

04/14/2026

Innodata is proud to join eMerge Americas as a Diamond Sponsor.

As AI continues to evolve, the conversation is shifting beyond models toward what it takes to build systems that actually perform in real-world environments. Data quality, evaluation, and infrastructure are becoming the defining factors.

Join Us in Miami, FL | April 22-24, 2026
https://emergeamericas.com/

04/14/2026

AI can count aircraft from space.
But understanding them is a very different problem.

In our latest article, Innodata’s Frank Tanner explores what happens when leading multimodal AI models are applied to satellite imagery, and where they fall short. While models perform well at counting objects, they struggle with classification tasks that require real-world context like scale, geometry, and physics.

The takeaway: today’s AI is powerful, but it still lacks something fundamental — a true “world model.”

This gap has real implications for defense, geospatial intelligence, and any application where physical understanding matters.

Read the full blog to see what this means for the future of AI:

AI can count aircraft in satellite imagery, but understanding them requires more. Learn why world models are the next step in geospatial AI.