IBM and University of Illinois propose a Ctrl-Z feature for agents that could reduce IT outage costs by $14,000/minute.
Read more
AI NewsEdge ComputingEnergy Efficiency
The future of AI is in your hands
IBM's Granite 4.0 Nano handles 88.7% of daily queries locally, cutting energy use by avoiding cloud calls.
Read more
AI NewsStorageAI Inference
Accelerating AI inference with IBM Storage Scale
IBM Storage Scale reduces time-to-first-token (TTFT) by 8-12x for LLM inference by providing a high-performance KV cache tier.
Read more
AI NewsAI HardwareOpen Source
Making open infrastructure for AI a reality, together
IBM Research and partners are transforming the AI age with complete solutions in an open ecosystem, launching the Spyre AI accelerator.
Read more
AI NewsGeospatialDisaster Response
IBM and ESA Release ImpactMesh Dataset to Enhance Flood and Wildfire Mapping
IBM and ESA released ImpactMesh, a novel multi-modal dataset, improving burn scar map accuracy by at least 5%.
Read more
AI NewsLanguage ModelsIndia Tech
IBM Granite 4.0: Hyper-efficient, high performance hybrid models for India
IBM’s Granite 4.0 models cut GPU costs by 50% for Indian languages using hybrid Mamba/transformer architecture, certified under ISO 42001.
Read more
AI NewsEnterprise AIBenchmarking
IBM and Kaggle launch enterprise AI leaderboards for real-world benchmarks
IBM and Kaggle introduce leaderboards to standardize AI model evaluation for complex enterprise tasks like IT automation and asset management.
Read more
AI NewsLLMsAI Architecture
Teaching LLMs to Count: IBM's PD-SSM Breakthrough
IBM's PD-SSM model achieves 98.5% accuracy on state tracking tasks, addressing LLM limitations in sequential reasoning.
Read more
AI NewsSoftware EngineeringArtificial Intelligence
IBM’s Software Engineering Agent Tops Leaderboard for Java
IBM’s iSWE-Agent achieved 33% success in resolving Java issues on the Multi-SWE-Bench, surpassing previous leaderboard holders.
Read more
AI NewsSoftware EngineeringAutomation
Teams of agents can take the headaches — and potential costs — out of finding IT bugs
IBM’s Project ALICE, a multi-agent system, demonstrates a 10-25% improvement in identifying root causes of IT issues.
Read more
AI NewsTransparencyLLMs
IBM Granite is Ranked World’s Most Transparent Model
IBM Granite achieved a 95% score on the Stanford Foundation Model Transparency Index, surpassing all other models by 23 percentage points.
Read more
AI NewsAgent AIDevOps
ToolOps: Enhancing Tool Reliability for AI Agents
IBM Research introduces ToolOps, a set of ALTK components improving correct tool invocations by up to 10%.
Read more
AI NewsLLMsAI Evaluation
IBM and Notre Dame Open-Source Benchmark Cards for LLMs
IBM and University of Notre Dame released 105 validated benchmark cards and a dataset of 4,000 cards to improve LLM evaluation transparency.
Read more
AI NewsPhysicsMaterials Science
A new advance in a two-century pursuit in physics
A new method for characterizing semiconductor materials, leveraging a mathematical concept from 300 BCE, unlocks over 20 material parameters from a single measurement.
Read more
AI NewsSemiconductorsMachine Learning
Thermonat Models Heat with Unprecedented Accuracy
IBM’s Thermonat project achieved semiconductor heat prediction accuracy within 1°C, 50,000x faster than existing methods.
Read more
AI NewsQuantum ComputingPhysics
Can quantum computers model nature’s most turbulent systems?
New research from IBM demonstrates a potentially exponential speedup for simulating stochastic quadratic differential equations, a key step towards modeling complex turbulent systems.
Read more
AI NewsQuantum ComputingSupercomputing
Quantum-Centric Supercomputing with CPUs, GPUs, and QPUs
IBM researchers demonstrate quantum-centric supercomputing by combining CPUs, GPUs, and QPUs, achieving 100x speedup in chemistry simulations
Read more
AI NewsSelf-HostingData Privacy
Maybe we don't need a server
Explores the potential for serverless personal tech by syncing data as files, reducing reliance on centralized services.
ServiceNow-AI’s Apriel-1.6-15b-Thinker achieves state-of-the-art performance against models ten times its size, scoring 57 on the Artificial Analysis Index.