The new version adds image input support and an experimental long-context mode for processing documents and logs.
Visko's Orbis model achieves 1.07x playback speed on a single Cloud TPU slice using JAX and custom kernels.
The Gemini agent is not yet publicly available, but startups can contact sales to prepare for its launch.
The new model supports text-to-image, multi-reference editing, and bounding-box layout control, with distilled checkpoints generating 1024x1024 images in ~2 seconds on H100 GPUs.
The inner display is 50% larger than the iPhone 18 Pro Max and features the A20 Pro chip
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.
Converted the credit service's monolithic structure into a DDD-based architecture through 435 PRs
The model generates structured task and behavior descriptions from conversations to condition separate LLM-based user simulators, rather than generating user utterances directly.
MASS based on Qwen3.6-27B improves research benchmark performance by 1.2–1.6x, while SW benchmarks decline
A rank-16 LoRA distilled from representation analysis cuts the slop score by 83% without system prompts or significant quality loss.
The update introduces opt-in 'danger zone' flags for image support, a new 'box' type, and extended context lengths, alongside a new model version focused on multilingual improvements.
The new Sabi-Noema models reduce token generation by 50% on TB2.1 while outperforming GPT 6 Luna-medium on DeepSWE benchmarks.
The updated benchmark reveals that only 3 of 16 tested LLMs maintained over 90% agreement across five test runs.
Analysis reveals malicious traffic shifted to other Russian hosting providers like Radist and Selectel while search indexing collapsed.
Netflix experimented with embeddings, subgraph integration, and fine-tuning to balance latency, cost, and model power in production.
By October 7, 120 of 160 production services were deployed in Central US, and Azure handled 65.4% of read traffic for GitHub’s core application.
DataJunction reduces metric onboarding time from weeks to minutes by providing a centralized semantic layer with SQL parsing and generation capabilities.
Netflix researchers developed a hybrid approach combining large-scale behavioral modeling with small-scale A/B tests to measure the causal impact of personalized recommendations, finding that targeting effects significantly outweigh mechanical exposure.
The new Cassandra-based cold tier stores over 15 PB of data and reduces disk usage from 80% to 7% while maintaining sub-500ms p99 latency.
The new Flink-based pipeline processes hundreds of thousands of events per second during live events with negligible errors, eliminating single points of failure in ad serving.
Netflix has rebuilt its Conductor workflow engine to support up to 30,000 tasks per workflow and handle approximately 420 million monthly executions, achieving a 40% reduction in p99 evaluation latency.
The engineer reduced multi-hour sequential debugging cycles to parallel runs by isolating experiments via Git branches and automating deployment with AI agents.
Netflix's system uses Reasoning-Aligned Rubric Tuning and weekly human reviews to maintain judge accuracy across millions of monthly explanations.
The system automatically adjusts CPU, memory, disk, and network limits for high-throughput media workloads, reducing manual tuning overhead and infrastructure costs.
AI Briefing