AI BriefingAI21 Labs news
The latest AI and developer news from AI21 Labs, with short summaries.
Visit website →AI21 Labs Releases Teel, an Open-Source Python Library for Parallelizing Agent IO
The library uses a shadow interpreter to parallelize independent IO-bound calls in sequential code without requiring async syntax, reducing runtime from 148s to 24s in benchmarks.
AI21 and Deel post-train Nemotron 3.5 Super VL to outperform Claude Opus 5.5 on global payroll at 50x lower cost
A three-week engagement used near-policy SFT and SDPO to lift the model's quality by 20.7 points over the base and 8 points over the frontier competitor.
AI21 Builds LIFT-Bench to Fix Instruction-Following Training Collapse
Standard instruction-following rewards cause models to collapse to short, empty answers; LIFT-Bench fixes this by scoring content alongside formatting constraints.
AI21 cuts GitLab Duo Code Review cost by 98% via harness optimization
The 98% saving required combining a switch to DeepSeek V4 Flash with context compaction to fit oversized reviews within the model's deployment window.
AI21 Post-Trains NVIDIA Nemotron 3.5 Super VL with Gong, Deel, and Alice for Specialized Enterprise Models
AI21’s Forward-Deployed Researchers post-trained NVIDIA Nemotron 3.5 Super VL with three partners in three weeks, achieving frontier-level performance at a fraction of the cost.
Better and cheaper together: Open models explore, frontier models patch
AI21 analysis of internal traffic reveals public coding benchmarks overestimate task difficulty, leading to inefficient model routing.
AI21 Proposes Verifiability Litmus Test to Guide Agent Architecture
AI21 argues that an agent's ability to verify its own outcome should determine whether resources are spent on verification or on diversity and aggregation.
AI21 Cuts High-Priority GPU Time-to-Start by 83% Using Kueue on Google Cloud
AI21 eliminated 20 weekly manual interventions and reduced hero job starvation from 72 to 12 hours by implementing Kueue on a 10,000-GPU cluster.
A Verifier Is Needed
Introducing an independent verifier improved the correct-answer selection rate for agentic search, surpassing the previous SOTA.
PickCheaper, More Powerful Patch Generation Through Collaboration Between Open Models and Frontier Models
By using a tiered architecture where a cheap model extracts context and a frontier model writes the final patch, the approach captures both cost efficiency and performance at the same time.
Improving Best-of-N Performance via Budget-Aware Execution for SWE Agents
AI21 proposes a strategy that optimizes cost and speed for SWE agents by flexibly allocating compute resources according to task difficulty.
Costs Aren't Coming Down. So What Now?
As AI agents' token usage surges, automated routing technology that optimizes costs without sacrificing quality is becoming critical.
Combining Weak Agents to Build a State-of-the-Art Deep Researcher
Instead of developing a new agent, merging the outputs of existing agents achieved the #1 spot on the Deep Research benchmark.
Scale First, Enrich Later: Achieving SOTA on SWE-rebench with the Right Execution Strategy
AI21 achieved SOTA on SWE-rebench with a 60.9% resolve rate by flipping the execution strategy to 'scale first, enrich later.'
Caching for Agentic LLM Pipelines
A cache key design was proposed that satisfies both reproducibility and variability together in a non-deterministic LLM environment.
Maestro Deep Research Agent
AI21 Maestro achieved SOTA on BrowseComp-Plus with 95.18% accuracy.
Benchmark distortion caused by gold-like answers
The LLM judge favored patches that 'looked like gold,' distorting benchmark scores.
Pay Attention to the Gaps
AI21 presents four gaps—validation, context, latency, and decomposition—as the core challenges of production AI.
AI System Requirements
The bottleneck in AI development is shifting from coding to the process of translating human intent and context into clear specifications.
Key Challenges and Solutions for Enterprise Generative AI
This piece examines the major challenges that arise when adopting generative AI for enterprise use and analyzes ways to resolve them to achieve operational efficiency.