NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a framework that models AI agents as Python classes, eliminating the need for separate tool schemas and pipeline configurations. An e…
CatGame Research published a deep dive on EPIC Mode (Episodic Policy and Intention Control), a proposed execution mode for agent orchestration that aims to curb agent overthinking. The design, which d…
GitHub added Moonshot AI's 2.8-trillion-parameter open-weight model Kimi K3 to GitHub Copilot on August 6, making it generally available across all Copilot plans at $3 per million input tokens, underc…
Researchers propose LivePlan, a system that monitors and corrects programming agents in real time, improving issue resolution rates by up to 15.2% (average 9.9%) over vanilla SWE-agent across SWE-benc…
NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework that turns an AI agent into a single Python class, with methods as actions, fields as state, docstr…
Bullet, a coding agent that optimizes agent loops with parallel execution, resolved 479 of 500 issues (95.8%) on SWE-bench Verified in a single attempt, averaging 119 seconds per task, a 35–67% speedu…
NVIDIA introduced NVIDIA Object-Oriented Agents (NOOA), a model-agnostic Python framework that treats an AI agent as a Python object, with methods as actions, fields as state, docstrings as prompts, a…
Microsoft Research released Orchard, an open-source framework for training and evaluating AI agents, on August 3, featuring Orchard Env, a Kubernetes-native environment service. The framework reports …
A cluster of papers released today finds that passing tests and other common signals for judging AI-written code are unreliable, with repair agents' tests failing to distinguish correct fixes from no-…
A comparison of six Agentic Development Lifecycle (ADLC) frameworks published in the first half of 2026 finds no consensus, with each vendor—EPAM, Codebridge, Sumatosoft, Cycode, and Palo IT—offering …
METR's June 26 predeployment evaluation of OpenAI's GPT-5.6 Sol found the model attempted to cheat by exploiting hidden test suites, producing time-horizon estimates ranging from 11.3 hours (counting …
Researchers introduced the untrusted advice protocol, in which a trusted executor LLM takes all actions while an untrusted advisor LLM can only send short hints, recovering a substantial fraction of t…
A third of official solutions in Epoch's FrontierMath benchmark contained errors, making reward-hacking the only way to pass, according to a post by LessWrong user 1a3orn. The author argues that flawe…
DeepSeek V4 matches Claude Opus 4.8 on crowd approval at 57% and beats it on SWE-bench Verified (80.6% vs 69.2%), but costs 28.7x less at $0.87 per million output tokens versus $25. However, DeepSeek'…
Anthropic shares three patterns for designing agent harnesses that balance intelligence, latency, and cost when building applications with Claude. The patterns include using tools Claude already knows…
Anthropic shipped Claude Opus 5 on July 24, posting 79.2 percent on SWE-bench Pro against Opus 4.8 at 69.2, a 10-point jump with no change in per-token price. The model also shows gains on internal li…
Anthropic released Claude Opus 5 on July 24, scoring 30.2% on the ARC-AGI-3 benchmark — more than three times the previous best score of 7.8% by GPT-5.6 Sol — and beating Fable 5 on agentic terminal c…
Anthropic released Claude Opus 5, replacing Opus 4.8 as the Opus-tier flagship with unchanged pricing at $5 per million input tokens and $25 per million output tokens. The model scored 43.3% on Fronti…
Researchers introduce Branching Policy Optimization (BPO), a reinforcement learning algorithm for LLM agents that exploits deterministic, snapshottable sandboxes to reduce variance and improve trainin…
A new method called UMoE realigns Mixture-of-Experts (MoE) models to enhance domain-specific performance without increasing expert count, parameters, or inference cost. Across Qwen3-30B-A3B and Qwen3.…