AI Agents
AI Agent technology and applications
Articles
Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Model⭐7
Cognition's SWE-2 coding model, post-trained from Kimi K3, hits 50.0% on FrontierCode 1.1, matching Fable 5.1 at 64% lower cost and running only inside Devin.
Fly Language Model (FLM) Wires the Fruit Fly Connectome Into a⭐7
The Fly Language Model wires the fruit fly connectome into a frozen 1.2B LLM via a reservoir readout it trains on its own—yet its controls show the fly wiring d...
Can LLMs Engineer Their Own Agent Harness? ByteDance Seed's⭐9
ByteDance Seed's HarnessDev tests whether LLMs can build and evolve their own agent harness, finding only 34 of 64 changes generalize to unseen tasks.
Anthropic Adds Plugin Evals to Claude Code: 6 Grader Types, a No⭐7
Anthropic's `claude plugin eval` command for Claude Code runs plugin evals with six grader types, a no-plugin baseline, and a CI gate for skills.
Sakana AI Unveils Fugu Max and Fugu Ultra v2: Cheaper, Stronger⭐7
Sakana AI launches Fugu Max and Fugu Ultra v2, cheaper multi-agent orchestration models that balance capability and cost via a learned orchestrator API.
Google Research Releases ToolGrad: Answer-First Framework Hits⭐8
Google Research's ToolGrad flips tool-use data generation to answer-first, achieving a 99.8% pass rate and letting Gemma-3 match frontier models with just 500 s...
Redis LangCache: Managed Semantic Cache Cuts LLM API Costs by Up⭐8
Redis LangCache is a managed semantic cache that matches prompts by meaning, cutting LLM API costs up to 90% and returning cache hits 15x faster.
OpenAI Launches Agents API in Public Beta: One API Call Now⭐7
OpenAI's Agents API is now in public beta, letting developers run Codex-powered agents with one API call via managed sandboxes, sessions, and tools.
NVIDIA Details BioNeMo Inference Runtime (BioIR): 2.90x Higher⭐7
NVIDIA BioIR delivers 2.90x faster Boltz-2 folding and 58.5K residues per GPU-hour on 8xH100, scaling proteome-wide structure prediction in PyTorch.
DeepSeek-V4.1-Flash Launches with 1M Context, FP4 KV Cache, and⭐9
DeepSeek-V4.1-Flash launches with 1M context, FP4 KV cache, and cross-layer attention reuse—cutting KV memory to 890 bytes per token for input-heavy agent workl...