Benchmark: Latest AI News

InferenceX Puts GB300 First for MiniMax M2.7, but Only on a Fixed Chat Test

The new ranking gives infrastructure teams a comparable speed and cost snapshot, but its single-turn design does not test the multi-step agent work MiniMax highlights for M2.7.

2 min

Aug 31, 2026Models

Benchmark Models

ChatGPT o3 Handles Simulated Liquidity Choices, but Consistency Slips in Harder Cases

The simulated wholesale-payments exercise defines a narrow starting point for delegated cash management: routine payment priorities, with human oversight for potentially anomalous patterns.

2 min

Last week

5 stories

Aug 29, 2026Models

Benchmark Models

Google’s WikiSkill Lifts Agent Benchmarks by Giving Models a Memory of Failed Work

The framework turns task traces into reusable instructions instead of changing a model’s training, offering a practical route to more capable agents while leaving uneven task gains and cross-model transfer as constraints.

Aug 28, 2026Tools

Perplexity Search API Sweeps Artificial Analysis Test, but Medium Context Wins

The result gives developers a concrete retrieval-quality lead to test, while showing that sending an agent more page text is not automatically the best route to better answers.

Aug 25, 2026Infrastructure

OpenAI’s Jalapeño Claims 1.5–1.9x More AI Work Per Watt, Faces 2027 Scale Test

The custom chip gives OpenAI an early efficiency claim against available Nvidia systems, but production qualification, broader workloads and a small initial rollout will determine whether it becomes a meaningful serving platform.

4 min

Aug 24, 2026Models

Benchmark Models

NVIDIA Says AVO Took Claude Opus 5 From 30% to Perfect on ARC-AGI-3

The result shifts attention from training larger models to the software that gives them memory, tools and recovery loops. Its clearest limitation is equally important: the perfect score came on ARC-AGI-3’s public environments, while private sets remain the harder test.

Aug 24, 2026Infrastructure

NVIDIA Says Vera Rubin Delivers 30x More Agentic Throughput per Megawatt, Pending Review

The preliminary comparison shifts attention from fixed prompt benchmarks toward the power required to keep long, tool-using AI sessions responsive.

August 2026

5 stories

Aug 23, 2026Tools

InferenceX Adds 11 Telemetry Views to AgentX, Exposing What Benchmark Curves Hide

The new exploration layer makes cache setup, bursty subagents and queue behavior inspectable, but its best-curve design means readers must still examine each point’s underlying configuration before treating a curve as a like-for-like comparison.

Aug 22, 2026Tools

SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Agents

The claimed contribution is not a faster model kernel. It is a test workload that makes state retention, routing and data movement visible—and leaves open whether its fixes generalize beyond AgentX’s replay matrix.

Aug 22, 2026Tools

Aug 18, 2026Infrastructure

LinkedIn Measures AI Code Review Against Merged Code

The company’s reported acceptance rate is a useful outcome measure, but the more transferable design is a controllable review pipeline: local rules, pre-posting filters and infrastructure built for monitoring.

2 min

Aug 18, 2026Infrastructure

Nvidia Dynamo’s Agentic Speedups Target the Work Around the Model

The reported gains come from changing how serving systems route, retain and stream state under concurrency. The next challenge is whether these AgentX-specific improvements hold across production mixes and unfinished scheduling work.

4 min

AgentX Replays Claude Code Sessions to Test AI Serving Systems

The replay dataset is designed to measure infrastructure behind agentic coding workloads, not whether a model writes better code. Its value depends on whether synthetic traces preserve the traffic patterns operators need to serve.