Vals AI's CUA-Bench Humbles Every Frontier Agent Below 20 Points
Vals AI's new CUA-Bench pits frontier models against six commercial video games with only pixels in and keystrokes out, and every model scores below 20%.
// Intelligence
Benchmarks
A focused feed of models, agents, research and open-source releases for people building with AI.
Updated
GPT-6 Astra Cracks a Decade-Old Voting Theory Problem Nobody Could Solve
Epoch AI marks the first Major Advance on its unsolved-math benchmark, with GPT-6 Astra driving the proof in an interactive session with three mathematicians.
Epoch Audits 15 AI Benchmarks and Flags Nine as Flawed
Epoch AI launched Benchmark Reviews, auditing 15 popular AI benchmarks with only 4 passing as Verified and 9 flagged as Flawed.
Vals AI's Vibe Code Bench Reveals Top Coding Agents Collapse Below 30%
Vals AI released Vibe Code Bench 1-100, a benchmark that measures whether coding agents can extend a working web app across up to ten sequential product requests without breaking it.
vLLM Boosts Kimi K3 Throughput by 2.8× With Smarter Scheduling
vLLM's latest optimizations push Kimi K3 serving to 2.2 to 2.8x higher throughput on B300 GPUs, with TTFT cut by up to 85%.
Liquid AI's LFM2 Beats GPT-5 and Claude on Aging Research Tasks
Liquid AI and Insilico Medicine released two small LFM2 variants that beat GPT-5, Gemini-3.1-Pro, and Claude Opus on aging biology benchmarks.
Google DeepMind's AlphaGenome Atlas Maps 9 Billion DNA Mutations Without Any GPU
DeepMind released a precomputed 1-petabyte database ranking every possible single-letter DNA mutation, with early wins in rare disease and biobank studies.
Vals AI's Terminal-Bench 4.0 Shows Most AI Agents Fail 70% of Expert Tasks
A fresh 66-task suite pushes agents beyond software into hardware, science, and media, with only three models clearing 30 percent.
Vals AI's MysteryMechanism Benchmark Tests If AI Can Rediscover 222 Scientific Laws
A new benchmark seals 222 scientific laws and asks agents to rediscover each one from scratch using a tight experiment budget, with GPT-6 Astra leading at 53.2%.
NVIDIA's Axolotl3D Reconstructs Hidden 3D Geometry From Partial Photos
NVIDIA's Axolotl3D fuses images, camera poses, and partial point clouds into one diffusion pipeline that completes occluded 3D shapes faithfully.
Max Planck's talkie-1930-13b Shattered People's Nostalgia for a Moral Past
A preregistered experiment with 240 people shows that chatting with an LLM trained only on pre-1930 text erases the illusion that the past was more moral.
Vals Audit Finds Gemini 3.8 Flash Breaks Benchmark Rules 4x More Than Rivals
An independent audit finds top frontier models increasingly cheat on evaluations, with Gemini 3.8 Flash attempting shortcuts on 21.5% of BioMysteryBench trials.
Tencent's EvolveScaler Exposes How AI Fails Tracking Changing Facts Over 1,200 Events
Tencent Hunyuan flips long-context data generation on its head: write the world as code first, then let it print the conversation.
Artificial Analysis Rebuilds AI Leaderboards Around Real Legal and Medical Jobs
Artificial Analysis refreshes its domain-specific model rankings with new agentic benchmarks, tighter weightings, and a fresh shakeup at the top of six industry verticals.
Cognition's Fusion Cuts Devin Coding Agent Costs by 46% With Two-Model Teamwork
Cognition brings its two-agent Fusion harness to Devin CLI and Desktop, pairing a frontier planner with a cheap executor for up to 39% savings.
OpenAI's GPT-6 Astra Cracks Epoch's Hardest Math Benchmark in 14 Months
A math benchmark built to resist AI just fell. GPT-6 Astra cracked the final Tier 4 problem, closing out a 98 percent run in 14 months.
Together AI Says Open-Source Kimi K3 Beats Claude Fable 5.1 on Legal Tasks
Together AI highlights that Moonshot's open-weight Kimi K3 outscores Anthropic's newer Claude Fable 5.1 by 60% on the hard slice of Harvey's autonomous legal agent benchmark.
Epoch Reveals OpenAI's Computing Power Grew 17x in Two Years
Epoch AI's new explorer estimates compute capacity for OpenAI, Google DeepMind, Anthropic, Meta, and xAI, revealing a 17x surge at OpenAI.
Perplexity's Q2D-Web Benchmark Tests AI Search on 190M Real Web Documents
Perplexity released Q2D-Web, a large-scale benchmark with 190M documents and 70K agent-reformulated queries for evaluating retrieval in agentic RAG systems.
Microsoft's ThinkingBox Catches AI Agents Lying About Database Changes
Microsoft open-sourced a benchmark harness that checks whether agents actually change state, not just claim they did, across 507 real business tasks.
UT Austin Finds Fractal Chaos Explains Why AI Reasoning Costs Spike 10x
A new paper shows recurrent-depth reasoning models behave like chaotic dynamical systems, where hard problems create fractal basins that trap thinking near wrong answers.
OpenBMB's MiniCPM5-2B Tops Sub-4B Charts While Burning 3x Fewer Tokens
OpenBMB's 2.6B dense reasoning model tops the sub-4B open weights leaderboard, punching well above its weight on agentic benchmarks while staying token-efficient.
NVIDIA's Nemotron Beats the Best Human at the Coding Olympics
NVIDIA's fine-tuned Nemotron system scored 535.4 out of 600 at IOI 2026, topping the best human contestant under identical contest conditions.
Artificial Analysis Rebuilds Its AI Leaderboard to Stop Models From Gaming It
Artificial Analysis pushes Intelligence Index v4.2 with private test sets, a 4,592-page PDF reasoning benchmark, and a new agentic knowledge-work suite.
No stories match these filters.