Cohere published techniques to improve fairness in LLM serving (inference). The composition is strongly academic—one official Cohere source and four arXiv papers—so this reads less as a product launch than a research-backed argument about inference quality. It addresses correcting latency and resource-allocation skew when handling concurrent requests to keep service fair across users—an operational but quietly fundamental theme. The through-line is a shift of attention from scaling models up to serving the same model fairly and efficiently. It's still research-led; real-service impact and adoption are yet to confirm.
Inference × New Model Releases
Cohere tackles LLM serving fairness
Cohere tackles LLM serving fairness
✎ Story body
▲ Official & Press
Official
LLM Serving Fairness: No more noisy neighbours
Cohere ensures fair compute sharing across LLM serving tenants
Academic (arxiv etc.) 21 ▾
Academic
Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States
LOCUS releases a US local-ordinance corpus for legal AI
Academic
UBP2: Uncertainty-Balanced Preference Planning for Efficient Preference-based Reinforcement Learning
UBP2: uncertainty-balanced planning for efficient preference-based RL
Academic
IndicContextEval: audio-LLM context use across 8 Indic languages
Academic
A Technical Taxonomy of LLM Agent Communication Protocols
A technical taxonomy of LLM agent communication protocols
Academic
Towards an Agent-First Web: Redesigning the Web for AI Agents
Towards an agent-first web: redesigning the web for AI agents
Academic
Which paper sections best reveal research methods?
Academic
Improving Medical Communication using Rubric-Guided Counterfactual Recommendations
Rubric-guided counterfactual recommendations for medical communication
Academic
RubricsTree: scalable open-ended evaluation of personal health agents
Academic
Descriptor: Certus Caliber Classification Gunshot Dataset (C3GD)
C3GD: a public field-collected gunshot muzzle-blast sound dataset
Academic
Knowledge Reutilization in Meta-Reinforcement Learning
A meta-knowledge reutilization framework for meta-RL across agents
Academic
Unintended Effects of Geographic Conditioning in Large Language Models
Unintended regional biases from geographic conditioning in LLMs
Academic
Learning Fair Pareto-Optimal Policies in Multi-Objective Reinforcement Learning
Learning fair Pareto-optimal policies in multi-objective RL
Academic
Meta-classification of one-class models via ranking correlation and kNN
Academic
WallZero: Mastering the Game of WallGo with Strategic Analysis
WallZero masters the board game WallGo with strategic analysis
Academic
When Multiple Scripts Matter: Evaluating ASR in Clinical Settings
Evaluating ASR in clinical settings when multiple scripts matter
Academic
Beyond Domains: Reusing Web Skills via Transferable Interaction Patterns
Reusing web skills via transferable interaction patterns
Academic
Benchmarking LLM Agents on Meta-Analysis Articles from Nature Portfolio
A benchmark for LLM agents on Nature Portfolio meta-analyses
Academic
Phase, not magnitude, carries identity inside image classifiers
Academic
RAID: Semantic Graph Diffusion for True Cold-Start and Cross-Lingual Forecasting
RAID: retrieval-augmented diffusion for cold-start, cross-lingual forecasting
Academic
Benchmark suite for federated noisy-label medical image segmentation
Academic
FraudSMSWalker: Benchmarking Agentic Large Language Models for SMS-to-Webpage Fraud Detection
FraudSMSWalker benchmark targets URL-masked SMS-to-webpage fraud