AI21 Labs

AI21 Labs

http://www.ai21.com
https://www.linkedin.com/company/ai21/

Privately owned AI21 Labs, headquartered in the US, was founded in 2017 and employs approximately 230 individuals. The company specializes in software development, providing access to language models that power text generation and comprehension features.

Revenue

Founded

2017

Headcount

227

Headquarters

US

Primary Segment

Software Development

Ownership

Privately Owned

News Summary:

On March 3, 2026, AI21 Labs announced its participation in AI Dev Day 26 x SF, a gathering on the West Coast that will feature workshops, demonstrations, and a startup track focusing on emerging AI companies. Earlier the same day, the company also confirmed its attendance at NVIDIA GTC 2026, an event highlighting advancements in physical AI, AI factories, agentic AI, and large-scale inference, offering sessions, training, and networking opportunities.

Similar Companies

Subscribe for full access to AI21 Labs's profile

AI21 Labs offers products in the enterprise tech industry. AI21 Labs's product portfolio comprises of generative AI.

Subscribe for full access to AI21 Labs's products in full detail

Subscribe for full access to AI21 Labs's revenue in full detail

Apr

14th

2026

06:07

Artificial Intelligence, B2B Tag

Like Answers Reveal LLM Judge Bias in Coding Benchmarks

TL;DR While benchmarking our coding agent, we noticed our LLM judge — the component in Maestro that selects the best output from parallel agent runs — was performing suspiciously well. Contamination was the obvious suspect — but when we reproduced the same bias on a clean dataset the model couldn’t have seen, we realized something else was going on: the judge had learned to favor solutions that look like gold answers — minimal, clean, focused — over outputs that actually work.

Apr

5th

2026

02:43

Artificial Intelligence

Engineering the subconscious: Why Claude Code isn’t enough to build AI systems

TL;DR The bottleneck in AI development has shifted. It is no longer about translating natural language into code; it is about translating the human subconscious into a conscious specification. The end of software development For decades, we thought coding was the hard part.

Mar

17th

2026

05:50

Artificial Intelligence

TL;DR AI in production is won or lost in the gap between a promising model run and a production AI system that can reliably validate, route, orchestrate, and decompose work under real-world constraints. Where production AI is actually won When I sat down with Barak Lenz, our CTO here at AI21 for a recent episode of YAAP, I tried to do what every host does.

Mar

10th

2026

06:12

Artificial Intelligence, B2B Tag

Where enterprise AI deployments actually get stuck

Over the past few years, I have spent a lot of time with enterprise teams trying to turn promising AI use cases into real systems that support daily work. Early on, progress usually feels fast. A model performs well in testing.

Mar

3rd

2026

15:09

Events, Artificial Intelligence, B2B Tag

Meet AI21 Labs at AI Dev 26 x SF Join AI21 at AI Dev Day This flagship gathering returns to the West Coast with expanded workshops, cutting-edge demos, and a new startup track spotlighting emerging AI companies solving real-world challenges.

Mar

3rd

2026

14:44

Events, Artificial Intelligence, B2B Tag

AI21 Labs at AI Dev 25

Meet AI21 Labs at NVIDIA GTC 2026 From physical AI and AI factories to agentic AI and large-scale inference, GTC 2026 will highlight the breakthroughs shaping every industry. Join us for inspiring sessions, hands-on training, and opportunities to connect with experts and peers.

Feb

26th

2026

06:54

Artificial Intelligence, B2B Tag

Modular intelligence: a human-like model for agent orchestration

TL;DR Human language production can be instructive for building agent architectures that are similarly modular, with discrete stages for reasoning, planning, and execution and strategies that mirror external self-monitoring. Not only does this cognitive analogy more aptly reflect the increasing complexity of today’s agent flows, the modularity model produces AI systems that are more auditable, more diagnosable, and more amenable to principled improvement.

Feb

11th

2026

07:59

Artificial Intelligence, B2B Tag

Reducing LLM training waste with model-agnostic padding minimization

TL;DR: Padding is a significant source of wasted compute when training LLMs. This remains a problem for hybrid Transformer-SSM models, where methods like sequence packing are not easily applicable. We managed to eliminate ~90% of padding-related overhead by applying a model-agnostic approach that utilizes micro-batch-level truncation and padding-aware micro-batching.

Feb

5th

2026

09:20

Artificial Intelligence, B2B Tag

Go big or go OOM: the art of scaling vLLM

TL;DR Sharing LLM-as-a-Judge (JLM) deployments across multiple concurrent training jobs can reduce GPU underutilization, but makes the deployment vulnerable to buckling under load. We mitigated this by applying a two-pronged approach: optimizing single-node performance and scaling multi-node deployment, arriving at a strategy applicable to any high-throughput inference deployment facing variable load.

Jan

29th

2026

06:44

Artificial Intelligence, B2B Tag

One token to corrupt them all: a vLLM debugging tale

TL;DR While working on a new Jamba model, we noticed it would generate complete gibberish, but just once out of every thousand prompts. Finding and solving the bug sent us deep into the heart of vLLM, eventually touching on how its scheduler interacts with different model architectures.

Jan

29th

2026

03:05

Artificial Intelligence

Chunk size is query-dependent: a simple multi-scale approach to RAG retrieval

TL;DR Different queries need different chunk sizes, but RAG systems commit to one size upfront. In this blog, we show that indexing the same corpus at multiple chunk sizes (e.g., 100, 200, 500 tokens) and aggregating results with Reciprocal Rank Fusion (RRF) improves retrieval by 1–37% across benchmarks, without retraining models.

Jan

22nd

2026

12:52

Artificial Intelligence, B2B Tag

When sleeping in saves you money: dynamic data snoozing for efficient online RL

Tl;dr: When running GRPO on verifiable rewards, dynamic sampling does a lot to stabilize the training process. However, it may also slow training down dramatically. In this blog, we introduce “dynamic difficulty snoozing” as a simple and effective strategy for cutting compute waste without breaking training stability.

Analyst insight reports

Channel hopping: a hybrid approach to TV distribution

The trouble with live sports

What we learnt from NAB and what it means for IBC

Navigating radio automation and playout

Subscribe to access all analyst insight reports