AI Performance

A reported Argo-Bench result shows the leading AI agent solved 34.8% of 210 tasks, highlighting unresolved reliability limits for autonomous systems.

Anthropic launched Opus 5.5 with lower token prices, faster output and claimed Fable-level results, raising the bar for efficient enterprise AI.

NVIDIA says Vera Rubin NVL72 can deliver up to 30x more agentic inference throughput per megawatt than GB300, pending independent benchmark review.

Meta’s Muse Spark 1.3 improves agentic and coding scores, but its low per-task cost may matter more than its still-unfinished frontier challenge.

NVIDIA says Vera Rubin NVL72 can deliver up to 30x more agentic AI throughput per megawatt than GB300, targeting long-context inference costs.

Meta has launched Muse Spark 1.3, citing better coding and agentic-task performance as it competes with OpenAI and Anthropic.

NVIDIA says Vera Rubin NVL72 can deliver up to 30x more agentic AI throughput per megawatt than GB300, reshaping inference economics.

OpenAI says its Jalapeño chip improves AI inference speed and power efficiency across models, but the results remain vendor-reported ahead of deployment.

OpenAI unveiled Jalapeño inference benchmarks that beat Nvidia systems on speed and efficiency, but limited deployment and vendor-supplied data temper the claim.

NVIDIA says Vera Rubin NVL72 delivers up to 30x more agentic throughput per megawatt than GB300, reshaping inference economics for AI builders.

A Princeton and UC San Diego study finds AI agent skills improve execution more than knowledge, but retrieval collapses as libraries expand.

NVIDIA says its AVO agent reached a perfect ARC-AGI-3 score, highlighting how memory, supervision and tools can extend model performance.

NVIDIA released SkillEvaluator, an open-source framework that tests whether agent skills improve task results, safety, and efficiency across real runs.

OpenAI is previewing Ultrafast, an API tier that runs GPT-5.6 Sol up to 14 times faster, targeting real-time enterprise workflows with Cerebras.

Nvidia’s Nemotron 3.5 Lightning uses sparse activation and quantization to deliver fast open-weight inference, targeting high-volume AI agents over peak scores.

Oxford has published a live benchmark for AI information-operations risk, giving model makers and buyers a new safety signal to track over time.

OpenAI says two API settings tripled GPT-5.6 scores on ARC-AGI-3, underscoring how inference configuration can reshape model performance.

Anthropic’s Claude Opus 5 set a new ARC-AGI-3 high score, raising fresh questions about real reasoning gains versus benchmark targeting.

Germany’s Soofi S open model claims top fully open English and German benchmark scores, while its creators also disclosed and corrected a test-data leak.

NVIDIA says LangChain-tuned Nemotron 3 Ultra reached top open-model agent benchmark results, highlighting lower-cost enterprise AI stacks.