Videos

The next stage of AI agent development is not better models or harnesses but 'agent recipes' that encode a maker's...

Genpark's platform, combining design and coding with GPT Astra, lets you build a complete interactive...

GrokBot dropped to $20 and added features that let it connect to any service without an API, control other apps, and...

MiMo-V3 introduces two techniques—KV bridging and KV reuse—that shrink a million-token KV cache from 12.09 GB to...

Anthropic's Opus 5.5 rewrote HAProxy in Rust, passing nearly all of the original's regression tests in 9.5 hours at...

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

The PR bottleneck is getting worse because AI agents generate more code than humans can review. Matt Pocock argues...

Opus 5.5 (via Claude Code) combined with the free hyperframes tool produces one-shot video edits with generated...

The OpenRouter-Stripe acquisition is fundamentally a security story: the token economy is projected to reach $10...

Google's open-source Artemis tool connects AI coding assistants directly to Android devices via an MCP server,...

Claude Co-work, combined with a Notion task database, can be configured as a personal assistant that delivers a...

Spotify's Large Taste Model, used daily by a quarter of US premium subscribers, shifts personalization from ranking...

Recommendation systems follow the same power-law scaling curves as LLMs. Meta's research lead argues the field is...

Your coding agent's past conversations are a hidden dataset: every prompt, tool call, and thought is saved as JSONL...

DoorDash treats LLMs as offline reasoning engines for search and personalization rather than serving them online....

Claude Opus 5.5 delivers Fable 5.1-level performance at 40% lower cost than Opus 5, making frontier AI more...

Researchers identified two internal signals in language models—value neurons predicting success probability and...

Unreal Labs' Unreal Agent claims up to 40% lower inference cost than Codex on Terminal Bench 4.0 by changing the...

Opus 5.5's effort levels produce dramatically different results for the same complex task, with Extra delivering the...

Opus 5.5 can generate animated graphics, logos, and product demos from a single text prompt, producing results that...

Jev is a new AI that makes thousands of fast, cheap decisions using structured JSON inputs, distinct from text...

Google's Gemini 3.8 Flash TTS models finally ship voice cloning, but with heavy guardrails: a spoken consent clip...

Connecting Claude Opus 5.5 to Higgsfield lets you go from a single text prompt to a generated product image, a...

Skydio's system lets a single operator command multiple drones across the country, with autonomous launch, tracking,...

DeepSeek V4.1 Flash replicated roughly 90% of a subway FPS game originally built by Claude Fable 5.1, using only the...

Dyna Robotics reports that its Dyna-1 manipulation models reached 99.4% success on a 24-hour napkin folding run by...

Opus 5.5 dominates cost efficiency over GPT-6 models at higher intelligence levels, but GPT-6 models lead in token...

Moonlake AI's world models prioritize causal, manipulable simulations over pretty pixels, using code-generated...

Robotics has made negligible progress for 70 years because it has always been approached as a hardware problem...

The real threat to AI evaluator independence is not direct funding from labs but donated compute, scope control, and...

Android's open source code is being released later than the Pixel builds, with developer APIs and security patches...

WebLLM allows developers to run AI models directly in the browser on the visitor's GPU, eliminating inference API...

Using a combination of models — a powerful frontier model for planning and review, and a cheaper open model for...

In a 12-use-case head-to-head, Opus 5.5 beat GPT-6 Astra 8-4, but Astra was 45% faster and 38% cheaper overall. The...

GPT-6 Sol is a competent but unremarkable model that rarely makes mistakes but also rarely impresses — it's...

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

Opus 5.5 outperforms GPT-6 Astra in design-focused tasks across multiple benchmarks, winning three out of five...

Claude Opus 5.5's 58 vs GPT-6 Astra's 53 on Artificial Analysis's intelligence index is a weak signal for coding,...

Sarvam's 3B parameter state-space vision model beats models 100x larger on document AI by using an SSM architecture...

Nvidia's Nemotron 3 diarization model (roughly 100 million parameters, 4 GB GPU minimum) handles up to eight...

Cody Menefee argues that the labor bottleneck in rotational grazing for grass-fed livestock can be solved by...

AI-generated silent UGC videos with text hooks are the core of a marketing strategy that apps like Lifta and Couple...

Building agents on document intelligence requires decomposing the parsing problem: use traditional computer vision...

A multi-agent system for short-form video moderation at Meta tackles two problems: modality misalignment (content...

Radical Numerics' new model Omni outperforms previous genomic language models on variant effect prediction,...

Using a pipeline of VLMs for labeling and judging, then training a small task-specific model like RFDTR, is cheaper,...