What changed in AI on 16 Sept 2026
1,336 events occurred on this UTC day across 3 sections — grouped when several documents describe one release, generated from the database, nothing editorial.
Dates are when events occurredoccurred_at = effective date when a source states one, else the observation time. Hover any date in the list for the observation time. `today` folds events sharing a group_key (one release seen in several documents) into one row.2 back-filled historical events excluded · see them on the timeline · include here
Price moves
Price moves 8
OpenRouter changed pricing for GLM 5.3 Flash: $0.075 in / $0.25 out per 1M tokens → $0.1 in / $0.3333 out per 1M tokens
$0.075 in / $0.25 out→$0.10 in / $0.333 outopenrouterOpenRouter changed pricing for DeepSeek-V4-Pro-0813: $0.66 in / $1.98 out per 1M tokens → $0.9834 in / $2.9502 out per 1M tokens
$0.66 in / $1.98 out→$0.983 in / $2.95 outopenrouterOpenRouter changed pricing for Nemotron 3 Ultra: $0.6 in / $2.4 out per 1M tokens → $0.625 in / $3.125 out per 1M tokens
$0.60 in / $2.4 out→$0.625 in / $3.13 outopenrouterOpenRouter changed pricing for Mistral Small 3.2 24B: $0.075 in / $0.2 out per 1M tokens → $0.09375 in / $0.25 out per 1M tokens
$0.075 in / $0.20 out→$0.094 in / $0.25 outopenrouterOpenRouter changed pricing for MiniMax M1: $0.55 in / $2.2 out per 1M tokens → $0.4 in / $2.2 out per 1M tokens
$0.55 in / $2.2 out→$0.40 in / $2.2 outopenrouterOpenRouter changed pricing for Qwen3 VL 30B A3B Instruct: $0.13 in / $0.52 out per 1M tokens → $0.15 in / $0.6 out per 1M tokens
$0.13 in / $0.52 out→$0.15 in / $0.60 outopenrouterOpenRouter changed pricing for Qwen3 VL 30B A3B Instruct: $0.15 in / $0.6 out per 1M tokens → $0.13 in / $0.52 out per 1M tokens
$0.15 in / $0.60 out→$0.13 in / $0.52 outopenrouterOpenRouter changed pricing for Kimi K2.7 Code: $0.71 in / $3.5 out per 1M tokens → $0.7062 in / $3.21 out per 1M tokens
$0.71 in / $3.5 out→$0.706 in / $3.21 outopenrouter
Model changes
Model changes 4
Qwen3 VL 30B A3B Instruct: max output tokens changed from 32768 to 16384
Max output32.8K tokens→16.4K tokensopenrouterQwen3 VL 30B A3B Instruct: max output tokens changed from 16384 to 32768
Max output16.4K tokens→32.8K tokensopenrouterDeepSeek V3.2: parameter count changed from 685396921376 to 685355329792
Parameters685.4B→685.4BhuggingfaceNemotron 3 Ultra: max output tokens changed from 182520 to 32768
Max output182.5K tokens→32.8K tokensopenrouter
Research
Research 500
New paper: Causal neural set filtering for online multi-target tracking
arxiv- New paperPaperManaging Action Preconditions in Neuro-Symbolic RL: Three Placement Strategies for Embodied Agents
New paper: Managing Action Preconditions in Neuro-Symbolic RL: Three Placement Strategies for Embodied Agents
arxiv New paper: OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning
arxivNew paper: Driver Behavior Estimation at Signalized Intersections Using a Physics-Constrained Decision-Conditioned Autoregressive Transformer
arxiv- New paperPaperHintMiner: Automatic Question Hints Mining From Q&A Web Posts with Language Model via Self-Supervised Learning
New paper: HintMiner: Automatic Question Hints Mining From Q&A Web Posts with Language Model via Self-Supervised Learning
arxiv New paper: POSPAN: Position-Constrained Span Masking for Language Model Pre-training
arxiv- New paperPaperSigned p-adic Residual Encodings of Finite-Domain All-Different Systems with a Sudoku Case Study
New paper: Signed p-adic Residual Encodings of Finite-Domain All-Different Systems with a Sudoku Case Study
arxiv - New paperPaperYou Don't Need To Train: Agentic Heuristic Learning Studio for Executable Human Activity Recognition
New paper: You Don't Need To Train: Agentic Heuristic Learning Studio for Executable Human Activity Recognition
arxiv - New paperPaperA panoramic aerodynamic performance prediction method for turbomachinery cascades using transformer-enhanced neural operator
New paper: A panoramic aerodynamic performance prediction method for turbomachinery cascades using transformer-enhanced neural operator
arxiv New paper: A Dynamic Aggregation Strategy Enhanced Efficient Global Optimization Algorithm for Solving High-Dimensional Turbomachinery Design Problems
arxiv- New paperPaperBeyond Distribution Matching: Semantics-Consistent Tabular Diffusion with Weak Semantic Priors
New paper: Beyond Distribution Matching: Semantics-Consistent Tabular Diffusion with Weak Semantic Priors
arxiv - New paperPaperSchema-Adaptive Action-Conditioned JEPA for Cross-Machine CNC Transfer under Partial Sensor Overlap
New paper: Schema-Adaptive Action-Conditioned JEPA for Cross-Machine CNC Transfer under Partial Sensor Overlap
arxiv New paper: Pseudo-Label Augmentation for Affect Sensing in Small Collaborative Groups
arxiv- New paperPaperDistilling Foundation Models for Agentic What-If Reasoning:Cost, Latency, and Governance in a Hybrid LLM+SLM Architecture
New paper: Distilling Foundation Models for Agentic What-If Reasoning:Cost, Latency, and Governance in a Hybrid LLM+SLM Architecture
arxiv New paper: Evaluating Open-Weight E-Commerce Agents with Environment-Grounded Verification
arxivNew paper: SWB-DM: A Calibrated Sliced-Wasserstein-Barycenter Aggregator with Delayed-Momentum Caching for Byzantine-Robust Federated Learning under Partial Participation
arxiv- New paperPaperA Decision-Support Audit Protocol for Supervision Drift in Proxy-Labeled Credit-Risk Prediction
New paper: A Decision-Support Audit Protocol for Supervision Drift in Proxy-Labeled Credit-Risk Prediction
arxiv New paper: LLMs as Master Forgers: Generating Synthetic Time Series Data for Manufacturing
arxivNew paper: Skeletal Prototypes on Iterative Nerve Expansions
arxivNew paper: Z-Loss Backward Geometry in Dense Output Heads and Sparse Routers
arxivNew paper: Anatomy of Associative Recall in Fixed-State Recurrences: A Matched-State Decomposition, an Interference Wall, and a Curriculum That Breaks It
arxivNew paper: Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration
arxiv- New paperPaperHow I learned to stop worrying and love StopGrads: Stationarity, Convergence, and a case study on Flow Map Learning
New paper: How I learned to stop worrying and love StopGrads: Stationarity, Convergence, and a case study on Flow Map Learning
arxiv New paper: Test-Time Unlearning via Sparse Autoencoder
arxiv- New paperPaperEfficient Reasoning Distillation: Small Video-Language Models via Synthetic CoT and Difficulty-Aware Fine-Tuning
New paper: Efficient Reasoning Distillation: Small Video-Language Models via Synthetic CoT and Difficulty-Aware Fine-Tuning
arxiv - New paperPaperThe record is part of the task: matched-record evaluation of text classifiers across maintenance, safety and recall reporting
New paper: The record is part of the task: matched-record evaluation of text classifiers across maintenance, safety and recall reporting
arxiv New paper: Scaling Laws for Physics-Aware ACOPF Surrogate Learning
arxivNew paper: Differentially Private Semantic Plans for Aggregate Insight Generation
arxiv- New paperPaperDrift Field Net: Learning Ocean Lagrangian advection fields from in-situ and satellite observations
New paper: Drift Field Net: Learning Ocean Lagrangian advection fields from in-situ and satellite observations
arxiv
First 30 of 500 — all in the feed →
Quiet today: Major releases · Benchmark moves · Open-weight releases · Deprecations · Provider changes · Hardware. Timeline →
Events that occurred on this UTC day (effective date when known, else observation date), excluding back-filled history and source-document changes. `today` groups events sharing a group_key (one release seen in several documents).