Full archive

Latest AI News

Search reporting on AI markets, models, products, companies, policy, and culture.

Yesterday

2 stories

InferenceX Ranks B300 First for Qwen3.5 at 15,815 Tokens per Second

The live ranking gives operators a matched speed-and-cost comparison for a long single-turn chat request, not a verdict on every production inference pattern.

2 min

YesterdayInfrastructure

Benchmark Infrastructure

InferenceX Ranks GB200 First for MiniMax M3, While MI355X Has Lower Token Cost

The live AgentX result gives operators a comparable speed-and-price snapshot at one responsiveness target, but the ranking can shift with serving software and does not settle the wider configuration trade-offs.

3 min

Earlier this week

1 story

Sep 2, 2026Infrastructure

Benchmark Infrastructure

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

Last week

2 stories

Aug 25, 2026Infrastructure

Benchmark Infrastructure

OpenAI’s Jalapeño Claims Up to 1.9x Efficiency Gain, but Leaves Nvidia Training Intact

The Broadcom co-developed inference processor is headed for deployment by year-end. Its published gains are meaningful, but they exclude model training and Nvidia’s Vera Rubin platform.

3 min

Aug 24, 2026Infrastructure

Benchmark Infrastructure

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.

3 min

August 2026

7 stories

Aug 23, 2026Tools

Benchmark Tools

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.

3 min

Aug 22, 2026Tools

Benchmark Tools

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.

3 min

Aug 21, 2026Models

Benchmark Models

Open Models Are Catching the Frontier Faster. Benchmark Scores Aren’t the Whole Contest.

SemiAnalysis sees the lag from closed-model breakthroughs to open-model benchmark parity shrinking across three AI eras. But a comparable model score still leaves a harder product question: who can turn capability into dependable agent work?

4 min

Aug 18, 2026Infrastructure

Benchmark Infrastructure

LMCache Reworks the Cache Plumbing That Can Stall Long-Running Agents

The changes address a specific failure mode in long, concurrent agent sessions: reusable model state can fill a shared pool before any request makes progress. The reported gains are promising, but they come from targeted validations rather than a broad production benchmark.

3 min

Aug 18, 2026Infrastructure

Benchmark Infrastructure

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

Aug 18, 2026Products

Launch Products

Cerebras Turns Up the Power on Its CS-4 Inference Rack

The new system aims to double CS-3 performance through faster clocks, denser racks and more cooling—not a new chip. Its 44 GB of SRAM per wafer remains a constraint as Cerebras pursues partnerships for split inference workloads.

3 min

Aug 18, 2026Infrastructure

Benchmark Infrastructure

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.

3 min

Superpower Daily

Your reading space

Latest AI News

Yesterday

InferenceX Ranks B300 First for Qwen3.5 at 15,815 Tokens per Second

InferenceX Ranks GB200 First for MiniMax M3, While MI355X Has Lower Token Cost

Earlier this week

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

Last week

OpenAI’s Jalapeño Claims Up to 1.9x Efficiency Gain, but Leaves Nvidia Training Intact

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

August 2026

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

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

Open Models Are Catching the Frontier Faster. Benchmark Scores Aren’t the Whole Contest.

LMCache Reworks the Cache Plumbing That Can Stall Long-Running Agents

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

Cerebras Turns Up the Power on Its CS-4 Inference Rack

AgentX Replays Claude Code Sessions to Test AI Serving Systems