LLMs & Models
Large language models: GPT, Claude, Gemini, Mistral and open weights.
97 published articles
Autonomous Agents
Hermes Agent gets smarter the longer it runs, and that changes everything
Hermes Agent from Nous Research goes beyond the typical coding copilot by embedding a closed learning loop, cross-session memory, and autonomous skill creation. It runs anywhere, integrates with 20+ messaging platforms, and aims to build a deepening model of each user over time.
2026-07-19
AI research
Alibaba's Qwen just turned world modeling into a language problem
Qwen-AgentWorld is a native language world model that simulates agent environments across seven domains. By making environment modeling the training objective from the start, it offers an alternative to video-based world models and domain-specific simulators for embodied AI.
2026-07-19
New benchmark exposes agent reasoning limits
AI agents hit 49% on a test human experts pass at 95%. More compute won't fix it.
Alibaba's HSCodeComp benchmark reveals the gap: top agents hit 49.4% accuracy vs. human experts' 95% on tariff classification. The bottleneck is structural, more compute doesn't help.
2026-07-19
Picbreeder revived with VLM agents
When AI agents rebuilt the internet's most creative experiment, they hit a wall humans never see
Sakana AI, MIT and NYU recreated Picbreeder with VLM agents to study open-ended creativity. The agents became trapped in repeating motifs, unable to make the conceptual leaps humans manage. Diverse personalities helped, but the gap between artificial and human creativity remains unexplained.
2026-07-19
AI Coding
Kimi K3 usage on Opencode doubled in one week, per the platform's own count
Opencode's X account reported that Kimi K3 usage doubled on the platform over the past week. The post doesn't give completion counts or name a competitor, so claims that K3 is now the most-used model should be treated as unconfirmed.
2026-07-18
Embodied AI
Mistral's tiny 8B model just made lidar optional for office robots
Mistral AI's Robostral Navigate is an 8B model that uses only one RGB camera to achieve 76.6% on R2R-CE benchmarks, outperforming multi-sensor approaches by 4.5 points. Built with simulated data and a token-efficient prefix-caching method, it generalizes across robot types and adapts to unseen obstacles.
2026-07-17
Multimodal AI
Mistral's 12B model just embarrassed a 90B one. The scaling orthodoxy has a problem.
Pixtral-12B matches or beats models seven times its size on multimodal benchmarks, without sacrificing language performance. Mistral also releases a new open benchmark for practical vision-language evaluation, challenging the idea that bigger is always better.
2026-07-17
LLMs & Models
Mistral's cascade recipe shrinks LLMs without killing reasoning
Mistral's cascade distillation shrinks large models into small ones while preserving reasoning and vision. The 3B variant packs capabilities that used to require ten times the parameters, and it's all Apache 2.0.
2026-07-17
benchmark breakdown
The 2.8 trillion parameter model that beats the frontier on the benchmarks that matter
Kimi K3, the 2.8T-parameter open model from Moonshot AI, trails frontier proprietary models on most broad benchmarks, but leads on SWE Marathon, Terminal-Bench 2.1, BrowseComp, and others. The detailed table reveals where its architectural bets on KDA and Stable LatentMoE pay off.
2026-07-17
Artificial Intelligence
Alibaba just open-sourced a world model that lets AI agents train inside a simulator
Alibaba's Qwen team released Qwen-AgentWorld, a language world model that simulates agent environments across seven domains. Its three-stage pipeline, CPT, SFT, RL, produces a simulator that beats GPT-5.4 on fidelity and enables agents to train via mental rehearsal rather than costly real-environment rollouts.
2026-07-16
Document AI
Mistral's OCR 4 scores big, but its own audit shows why benchmark numbers don't tell the real story
Mistral OCR 4 introduces bounding boxes, block classification, and confidence scores alongside text extraction, supporting 170 languages. It achieves 72% human preference win rates and top benchmark scores, but Mistral's own analysis shows standard benchmarks penalize correct output for formatting artifacts, not accuracy errors.
2026-07-16
Competitive programming
NousCoder-14B just opened the coding RL black box that OpenAI and DeepMind keep locked
Nous Research drops NousCoder-14B, a 14B competitive programming model with a fully open RL pipeline. The 68% Codeforces solve rate is notable, but the real story is that anyone can now replicate the stack.
2026-07-16