Agents

AICursor's Aman Sanger says the new model is an enormous improvement over Composer 2.5 and was trained entirely from scratch with the SpaceXAI team. The quoted Cursor post identifies it as Grok 4.5, Cursor's most powerful model yet and the first built for more than software engineering.

AICursor co-founder Michael Truell announced Grok 4.5, a model trained with SpaceXAI that the post calls Opus-class, fast, and low cost. He says it is a significant step up over Composer 2.5 and has become the daily driver for many on the Cursor team. A benchmark table shows Grok 4.5 at 83.3% on Terminal-Bench 2.1 and 78.0% on SWE-Bench Multilingual, with the post saying more releases will follow.

AICognition launched SWE-1.7, which it says reaches frontier-level coding performance at lower cost, trained from a Kimi K2.7 base. The post describes RL methods including top-p sampling replay to preserve entropy, compressed weight deltas across multi-cluster training, and self-compaction for rollouts up to six hours. SWE-1.7 is available in Devin via Cerebras at 1000 TPS.

Why it matters: The post details entropy preservation, multi-cluster weight sync, and self-compaction, offering concrete RL training techniques for long-horizon coding agents to compare against one's own pipeline.

AIBerkeley AI Research authors argue that near-free inference will make agents the dominant workload for data systems, requiring redesign for agentic speculation, agent-run state and coordination, and agent-synthesized systems. The post cites inference prices falling 9x to 900x per year with a median near 50x, and reports that about 80-90% of sub-queries in a text-to-SQL benchmark were duplicates. It frames the three directions as data systems for, of, and by agents.

Why it matters: The piece maps three concrete data-system challenges posed by near-free inference, useful for anyone designing infrastructure for agent workloads and memory.

AILilian Weng published a new post on harness engineering for AI self-improvement. She expects harnesses to evolve toward self-improvement and enable auto-research, while smarter models keep harnesses simple. Even if many harness gains are later internalized into core models, specifying goals and context will remain necessary.

AIMeta Superintelligence Labs has released Muse Image, which can invoke search and coding tools and self-refine its generations before output. It is available today in the Meta AI app, meta.ai, Instagram Stories in the US, and WhatsApp in limited countries, with Facebook coming soon. Meta also previewed Muse Video, which is coming soon to creators and Meta AI and is reported as ranking No. 3 on Arena for text-to-video at the time of writing.

Why it matters: The source describes how search, code execution, and self-refinement change image generation, which matters to anyone comparing agentic media models with plain prompt-to-image systems.

AITufa Labs won the first $37.5K ARC-AGI-3 milestone prize with "The Duck," a small open-source LLM that plays the games by writing and running Python in a live REPL. Reki placed second with a vision-language agent using Gemma-4-31B, and md Boktiar Mahbub Murad placed third with the "forge" framework. The second and final milestone prize ends September 30.

AIThe post argues that the system surrounding a base model, called the harness, increasingly determines how well AI agents deploy and improve. It reviews research where harness components such as workflows, context, and code are optimized automatically through evolutionary search and meta-agent loops. The author concludes that evaluators, memory management, and human oversight remain open bottlenecks.

AIAnthropic's Noah Zweben promoted a journey from Claude Code to Claude Tag, framing it as the future of working with AI. The related Claude post says Claude Fable 5 is now available in Claude Tag and features a conversation with Boris Cherny and Cat Wu on how the product spread from engineering across Anthropic.

AICognition launched the Devin Security Vulnerability Remediation Program, in which its forward-deployed engineers embed with customer teams to deploy Devin to find, validate, and fix vulnerabilities. The program first works through existing scanner backlogs from tools such as Snyk, SonarQube, and Semgrep, shipping validated fixes as pull requests, then adds Devin Security Swarm for continuous discovery of logic flaws. Most engagements run about six weeks, and eligibility is limited to enterprise Devin Cloud customers meeting the program's requirements.

AIPromptArmor reports that Skills in Microsoft Copilot Cowork can call DeepSeek even when an organization has not opted into the DeepSeek Preview. The calls use the agent's own access path, so users need no API key, and a Skill built this way received a 100/100 score from Microsoft's Skill Scanner. After Microsoft removed the DeepSeek Preview setting on June 25, the report says admins had no remaining setting to block DeepSeek through the Cowork code environment, leaving disabling Cowork entirely as the only option.

AICognition has launched Devin Security Swarm, which uses parallel agents to find vulnerabilities across a codebase, confirms exploitability in isolated sandboxes, and opens remediation PRs. In an evaluation on 50 real-world GitHub Security Advisory vulnerabilities, Devin reached 72% recall at about $90.23 per run, compared with 68% for Claude Security at $131.87 per run. The product is available starting today, with scan profiles and incremental scans that process only changed code after the first full baseline.

AIJim Fan announces ASPIRE, a system where coding agents use multimodal sensory traces from simulation and real robots to run evolutionary search over control programs and add the results to a growing skills library. The post claims up to a roughly 10x reduction in transfer learning tokens for sim-to-real and single-arm to bimanual transfer, and says the full stack will be open-sourced.

AIMollick argues AI capability is improving at a better-than-exponential rate, citing METR, GDPval, Epoch, and his own tests showing models working autonomously for hours. He says work is shifting from co-working with chatbots to assigning tasks to agents, with OpenAI workers managing multiple agents and experts getting the most from them. He adds that open-weights Chinese models trail the American frontier by roughly 6-12 months.

AICognition has released a preview of Devin Fusion, a multi-model harness that runs a frontier main agent alongside a cheaper sidekick agent. On FrontierCode 1.1 Extended, the company reports scores near frontier models at up to 60% lower cost per task, and 41% lower cost when paired with Fable 5, which access was suspended from June 12, 2026.

Why it matters: The post explains a sidekick architecture with cached persistent contexts, which contrasts with advisor-style tools and shows how cost cuts depend on the main model's delegation behavior.

AISebastian Raschka's tutorial shows how to build a fully local coding agent by pairing an open-weight LLM served through an inference runtime with an open-source harness that can read files, edit code, and run commands. He recommends Qwen-Code for Qwen3.6, citing Nvidia's Polar paper, which found Qwen models performed best in Qwen-Code. The Qwen3.6 35B-A3B model is about 22 GB to download and needs roughly 30–40 GB of RAM.

AIOpenAI's Economic Research team published a new paper on the shift from chat to delegation, where people hand longer, more complex work to agents. The post draws on Codex usage inside OpenAI as a preview of what agentic work may look like in the future.

AIAndrej Karpathy describes Claude Tag, which lets teams tag Claude into Slack channels and delegate tasks to it. He calls it a third paradigm for LLM interfaces, after websites and downloaded apps, with a persistent, asynchronous entity working alongside teams with org-wide tools and context.

AIEugene Yan says he likes Claude Tag's multiplayer form factor because other people can reply in the thread to give Claude context and direction. Claude Tag, per the background post, lets teams in Slack tag Claude in as a team member with access to chosen channels and tools.

AIPromptArmor demonstrated that OpenAI's Approve-for-me agent approved a malicious NPM install with elevated privileges after a hidden prompt injection in an external GitHub issue influenced the main Codex agent. The malicious package's post-install script then ran unsandboxed with the user's full privileges. The report also gives steps for organizations to disable agentic auto-review in Claude Code and Codex.

Why it matters: The report shows a prompt-injected GitHub issue leading an approval agent to permit a malicious NPM install, a concrete test of agent-in-the-loop guardrails.

AIOpenAI's alignment team found that re-generating 100,000 WildChat conversations with five recent OpenAI models predicted production failure rates across four orders of magnitude, with 95% of predictions within 1.04 orders of magnitude. The approach was weaker for agentic misalignment categories, where errors were about 37 times larger, and it still held roughly without access to chain-of-thought reasoning, with mean multiplicative error rising from 3.6x to 4.0x.

Why it matters: The post tests whether public chat data can predict real production failure rates, and where that prediction breaks down for agentic behavior.

AISpaceX has exercised its option to acquire Cursor in an all-stock transaction, aiming to build highly useful AI models. The quoted announcement says SpaceXAI has spent several months jointly training a model with Cursor, to be released soon in Cursor and Grok Build.

AIXiaomi has launched MiMo Claw, an agent built on its flagship MiMo model and integrated with Kingsoft Office for Word, Excel, PowerPoint, and PDF workflows. The company says it consumes 40–60% fewer tokens than comparable solutions, and daily usage has been expanded from 1 hour to 4 hours, with free access and no deployment required. A limited-time subscription is priced at ¥14.9 per month.

AISpaceX has exercised its option to acquire Cursor in an all-stock transaction, with the stated goal of building the world's most useful AI models. Over the past few months, SpaceXAI has been jointly training a model with Cursor, which will be released in Cursor and Grok Build soon.

AIZ.ai has released GLM-5.2, its flagship model for long-horizon tasks, which it says substantially improves on GLM-5.1 and supports a 1M-token context. The model adds IndexShare, which cuts per-token FLOPs by 2.9× at 1M context, and is released under the MIT open-source license.

Why it matters: The source gives benchmark tables against named rival models and deployment settings, useful for judging where GLM-5.2 sits among current flagship models.

AIZ.ai's release notes list GLM-5.2 as supporting 1M lossless context, with improved long-horizon task performance and reduced context drift and goal forgetting. The company says GLM-5.2 achieves open-source SOTA performance on coding and long-horizon task benchmarks. The page also includes the newer GLM-5.3 and GLM-5.3-Flash entries, which are listed above GLM-5.2.

Why it matters: The page lists a dated series of Z.ai model releases, showing how the coding and long-horizon agent line has evolved from GLM-4.5 through GLM-5.2.

AIMoonshot AI released Kimi K3 on Hugging Face as an open-weight, native multimodal agentic model with 2.8T total parameters and 104B activated parameters. It supports a 1-million-token context window and text and image input, with weights released under the Kimi K3 License. The model card reports benchmark results for coding, agentic, and vision tasks against several closed models, and recommends vLLM, SGLang, or TokenSpeed for inference.

Why it matters: The release pairs open weights with a 2.8T-parameter MoE architecture and benchmark tables against several named closed models, useful for comparing frontier capability claims.

AIOpenRouter introduced Fusion, a tool that sends a prompt to a panel of models and has a judge model fuse their results into one answer. On 100 DRACO deep research tasks, a Fable 5 and GPT-5.5 panel scored 69.0%, above Fable 5 alone at 65.3%, and a budget panel of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro reached 64.7% at about half the cost of Fable 5.

Why it matters: The source gives benchmark scores, panel compositions, and contamination controls, letting readers judge how much of the gain comes from model diversity versus self-synthesis.

AIMoonshot AI published Kimi-K2.7-Code, a coding-focused agentic model built on Kimi K2.6, with a 1T-parameter MoE architecture and 32B activated parameters. The model card reports about 30% fewer thinking tokens than K2.6 and benchmark results against GPT-5.5 and Claude Opus 4.8, with weights and code released under a Modified MIT License.

Why it matters: The model card gives benchmark comparisons against GPT-5.5 and Claude Opus 4.8 on coding and agentic tasks, useful for judging its position among current coding models.

AIZed is building DeltaDB, a version control system that records every operation as a fine-grained delta, linking agent conversations to the code they produce. The company says a beta will arrive in a few weeks, and it invites users to join a waitlist. The system is designed so teammates can collaborate on work in progress without waiting for commits, pull requests, or pushes.

AIXiaomi's MiMo team released MiMo Code, an MIT-licensed terminal coding agent built on OpenCode for long-horizon programming tasks. The design centers on three areas: Max Mode parallel sampling that generates five candidates per turn, Goal-based completion verification, and a memory system that checkpoints session state and rebuilds context. The article reports offline benchmark results and a double-blind A/B test with 1,213 pairs in which MiMo Code's win rate exceeded 65% beyond 200 execution steps.

Why it matters: The article explains how MiMo Code handles long-horizon coding through computation, checkpointed memory, and cross-session evolution, useful for judging design tradeoffs in coding agents.

AIEthan Mollick, who had early access to Claude 5 Fable, reports that it outperformed other public models in his tests, including an isochrone travel-time map and a nine-and-a-half-hour software build called Concord. He says the model delegated work to other agents and made many design choices he could not see or weigh in on, leaving him closer to a client than a hands-on operator. He also notes high token usage, frequent fallback to Claude 4.8 Opus under security guardrails, and persistent quirks in its writing style.