Topic

#Benchmarks

19 articles on Benchmarks — news, releases, guides and analysis from the SourceFeed engine.

One Vendor Finally Published the Attacks It Can't Catch

An open benchmark with a public failure list beats unverifiable detection claims — but a self-scored 99.8% still isn't proof.

Ji-ho Choi

The Red Queen Comes for Your Static Evals

Cambridge and NVIDIA's co-evolving Godel machine shows why fixed benchmarks are the real ceiling on self-improving agents.

Article · 1w ago0

Models Aren't Getting Dumber — They're Getting Unbundled

Active parameters collapsed while reasoning scores soared, and your retrieval stack quietly became responsible for the facts.

Article · 1w ago0

The New Frontier Models Are Amnesiacs by Design

Labs now trade stored facts for reasoning and bet retrieval fills the gap. Mostly it works, with one catch.

Article · 1w ago5

Z.ai Caught Up on Finding Bugs, Not on Exploiting Them

GLM-5.3's benchmark parity with Anthropic's restricted Mythos 5 hides an exploit-development gap that should reassure defenders.

Article · 2w ago3

Bullet Is Fast, but It's Built on Rented Land

The YC startup's latency pitch is the right fight; its bring-your-own-subscription funnel already got other harnesses banned.

Article · 2w ago2

Anthropic's New Benchmark Scores Reasoning Nobody Can Verify

The Conceptual Reasoning Index grades argument quality and logical coherence in domains with no ground truth.

Article · 2w ago5

There Is No Best Language for Coding Agents

Dan Luu's large-scale evals show the token-efficiency case for terse languages collapses on real tasks.

Article · 2w ago2

Intel's 18A Just Killed the ARM Efficiency Myth

A $699 Dell out-sips Apple's cheapest MacBook per watt — here's what that actually proves.

Article · 3w ago0

Claude Opus 5 Is Anthropic Undercutting Itself, on Purpose

Near-Fable-5 performance at unchanged Opus pricing rewrites the default model choice for agentic workloads.

Article · 1mo ago2

GPT-5.6 Sol Rewrites the Economics of Agentic Coding

OpenAI's new flagship matches Claude Fable 5 on intelligence for a fraction of the cost, but introduces new architectural trade-offs.

Article · 1mo ago0

Beyond Bug Fixing: The Rise of Senior-Level AI Coding Benchmarks

As coding agents max out early benchmarks, new evaluations target multi-file refactoring, non-contamination, and real-world ambiguity.

Article · 1mo ago2

GLM 5.2 Beats Claude on Cyber Benchmarks

An open-weight MoE model outscores Claude Code at finding IDORs, shifting the calculus for security tooling pipelines.

Article · 1mo ago2

The Open-Weights Gap Depends on What You Measure

A viral chart predicts open models reach parity by December 2026. Across 18 benchmarks, the honest answer is messier.

Article · 2mos ago5

Why GLM-5.2’s Low Hallucination Rate Upends the Enterprise LLM Stack

Z.ai’s MIT-licensed model challenges GPT-5.5 with a 3x lower hallucination rate and 1/7th the inference cost.

Article · 2mos ago2

GLM-5.2 Claims Top Open-Weights Spot on Artificial Analysis

Z ai's latest model matches proprietary frontier performance on agentic benchmarks while keeping an open MIT license.

Article · 2mos ago2