Videos

Meta's Muse Wonder models are competitive frontier models that score number one on the Deep Suite coding benchmark...

Meta's Muse Wonder models benchmark extremely high, but in hands-on coding and 3D game tests they underperform...

Meta's Muse Wonder is a competitive personal AI agent model that outperforms Opus 5.5 and GPT-6 Astra on legal and...

Local Qwen 3.8 on an RTX 4090 comes out to roughly 47 cents per million tokens of electricity, essentially tying...

Anthropic's Claude Haiku 5.5 is 90% cheaper than Haiku 4.5 for most prompts and shows major capability improvements,...

Gemini 4 Argon's 77.9% on Deep SWE would put it ahead of all other models, but the benchmark is public and...

Darwin 27B RSI, a model from Final Bench and Vidraft, claims a 5.24-point gain on GPQA Diamond through recursive...

Claude Sonnet 5.5 delivers a massive jump in agentic coding capability, scoring 70.6% compared to Sonnet 5's 10.3%,...

Mimo V2.6 Flash, the smaller 309B model, proved more reliable and capable than the trillion-parameter Pro in...

GPT-6 Luna is OpenAI's new cost-performance workhorse: 10 cents per million input tokens, 50 cents per million...

Claude Opus 5.5 is a cheaper, capable successor to Opus 5: Anthropic claims it matches a top-tier model (called...

Anthropic's Claude Opus 5.5 matches the performance of Fable 5.1 on most tasks while costing 40-51% less to run and...

Claude Opus 5.5 is now the frontier model, beating GPT-6 Astra and Fable 5.1 on key coding and knowledge-work...

Grok 4.7 is a cost-effective model that performs near the frontier, comparable to Opus 5 on Cursorbench at half the...

Neo Horse's post-training improved Qwen's agent performance without moving to a larger model, with the 4B model's...

DeepSeek V4.1 Flash is a 552B parameter MoE model with only 8B/16B active parameters, matching Opus 5 and GPT 5.6 on...

AI models do not think like humans; they lack the abductive reasoning that humans use to invent explanations from...

Replicating the IFScale benchmark shows the old 200-instruction ceiling is obsolete: today's frontier models hold...

GPT-6 Astra's 99.9% score on ARC AGI 3 came from OpenAI's own harness, while a neutral evaluation using Arc...

GPT-6 Astra, OpenAI's new frontier model, achieves near-perfect scores on ARC AGI 3 (99.9%) and Frontier Math...

Fable 5.1 tops Artificial Analysis's composite benchmark but scores only 67.4% on Deep Suite, placing it near...

Google's Gemini 3.8 Flash is a surprisingly capable small model that matches or beats frontier models on the Deep...

Fable 5.1 is the best model on benchmarks but costs more per task than Fable 5 because it uses 1.7x more output...

Claude Fable 5.1 keeps the same list prices as Fable 5 but is claimed to cost roughly 25% less in typical use and up...

James Zou presents a shift from designing AI workflows to designing environments for agents, illustrated by the...

Nvidia's NOOA (Object-Oriented Agents) framework argues that most agent framework machinery—prompt templates, tool...

Qwen released the 27B parameter Qwen3.8-27B model, which significantly outperforms its predecessor Qwen3.6-27B and...

This week four Chinese labs shipped frontier open-weight models, and Hugging Face's report shows the open-source...

GLM 5.3 is a 753B-parameter open-weight mixture-of-experts model that matches closed-source giants like Kimi K3 on...

Alibaba's Qwen 3.8-27B replaces 75% of its attention layers with Gated DeltaNet, slashing KV cache memory from 244...

Alibaba's Qwen 3.8 27B dense model claims to beat Opus 4.6 on several coding benchmarks while being Apache 2.0...

China's Z.AI released GLM 5.3, a post-training-only upgrade that found 2,436 vulnerabilities across 269 open-source...

Google's Gemini 3.7 Flash beats Claude Sonnet 5 on coding benchmarks at $0.75 per million tokens — half the price of...

Four new AI models hit the market in three days — Grok 4.6, DeepSeek V4 Pro, Nemotron 3.5 Lightning, and Muse...

Replay agents—blind scripts that replay recorded successful trajectories—can match or beat frontier models on...

DeepSeek shipped a new flagship model, an open-source agent harness, and a price list that raised one rate by 1,114%...

xAI dropped Grok 4.6, a point release that leapfrogs coding and knowledge-work benchmarks, making it a serious third...

16 GB VRAM just became the most common GPU config on Steam, which means the local AI tier list has shifted. Qwen 3.5...

Context windows are a lie: every major model degrades badly long before hitting its advertised limit, and the labs...

A new open-source model, Quen 3.8 Max from Alibaba, is competitive with top closed-source models like Fable and...

Alibaba released Qwen3.8 Max, a 2.4 trillion parameter mixture-of-experts model with 95 billion active parameters,...

Ling 3.0 Flash from Ant is a surprisingly competent coding model at 124B total/5.1B active parameters, outperforming...

Every company needs a private benchmark to reliably evaluate, release, and improve AI agents, moving beyond...

Claude Opus 5 is often cheaper than Fable 5 and can outperform it on coding and verification tasks, but Fable 5...

Jensen Huang and other AI leaders signed a letter supporting open-weight AI models, arguing open source drives...

Claude Opus 5 has been released with state-of-the-art performance on coding and knowledge work benchmarks, often...

Agent development is more like machine learning than traditional software engineering, requiring empirical...

Kimi K3 is the most powerful open-weight model released, but it suffers from reliability issues that public...

Kimi K3 is an open-source AI model from Moonshot AI that matches or beats closed-source models like Fable 5 and...

Sakana Fugu is an AI routing system that orchestrates multiple frontier models like Opus 4.8, Gemini 3.1 Pro, and...