#Deployment/Engineering

AIThe article explains how reasoning models can offer multiple effort modes, separating training-time methods from inference-time controls such as system prompts and chat templates. It compares six open-weight models, including DeepSeek V4, Nemotron 3 Ultra, Kimi K2.5, GLM-5, Qwen3, and Inkling, noting that their reports disclose different levels of detail. It also shows how GPT-5.6's model selection and effort settings act as two separate scaling axes.

AISekhar and Katie Brandt built Foreguard, a free tool built with ChatGPT and Codex, to help families claim public benefits they are entitled to and plan private insurance coverage. The tool shows that modest budgets of $100 a month can create millions of dollars of day-one financial protection. The goal is to help families prepare earlier, before care decisions become urgent.

AIDeepLearning.AI has launched a short course, built with Cerebras, on building LLM applications that respond quickly using inference-optimized hardware. The course compares how GPUs, TPUs, and Cerebras' Wafer-Scale Engine handle the memory-to-compute bottleneck, which keeps model weights close to compute units to speed token generation. It covers real-time applications such as live translation and voice agents, plus habits for agentic coding.

AIRun upgraded Gemma 4 with llama.cpp (QAT + MTP)

AIBunkerhill Health's Carebricks platform lets health systems create and deploy AI agents across clinical and operational use cases using data hospitals already generate. Sequoia Capital backed the company at seed and is continuing to invest. At UTMB Health, Bunkerhill grew from one agent in production to more than twenty, consolidating multiple vendors' point solutions into one platform.

AIModal has trained a DFlash speculator that runs much faster than MTP, according to Soumith Chintala. The speculator is backed by Inkling by Thinking Machines, which Modal says delivers 67% higher throughput and interactivity on Modal Auto Endpoints with SGLang.

AIThinking Machines introduced Inkling, an open-weight model with 975B parameters that reasons across text, image, and audio. The full weights are available, with fine-tuning on Tinker and access through the Inkling Playground and Hugging Face and partners.

AIVishal, who lost a leg at age six, used ChatGPT to plan his cycling training, nutrition, recovery, and prosthetic design research. After one year of competitive racing, he qualified for elite competition and earned a chance to represent India at the 2026 Asian Para Road Cycling Championships in Saudi Arabia and the Para Cycling Road World Cup in Thailand.

AICognition says its one-year-old merger with Windsurf has produced a more capable Devin, which now manages other Devins at a mid-to-senior engineering level, and new SWE-1.7 model, described as its most capable and efficient to date. The company reports growing from 44 to 350 people and revenue run rate from $73M to $500M+ since merging the brands.

AIArvind Narayanan's ICML keynote argues that AI's labor impact will depend on slow organizational adaptation rather than a single lab milestone. He cites reliability measurements showing agent accuracy rose much faster than reliability over the last 24 months, and points to software engineering and past technologies like electricity and ATMs. He concludes that evaluation work and human judgment will become more central as building tasks are increasingly automated.

AICognition found that Fable 5 led runs cost less than Opus 4.8 led runs on FrontierCode 1.1 when both used the same sidekick, $1.86 versus $2.04 per run. Fable 5 scored 60.7 against 54.6 for Opus 4.8 in those configurations, and it took fewer lead turns, delegated earlier, and rarely edited code itself. The post attributes the difference to delegation style rather than per-token price, and notes that the approach gives little benefit on short or serial debugging tasks.

Why it matters: The source compares lead-model delegation habits on a coding benchmark, showing how a pricier model can lower total agent cost through fewer turns and better handoffs.

AICognition's entire platform, including Devin Cloud, is now FedRAMP Class D (High) In-Process and listed on the FedRAMP Marketplace, extending FedRAMP High authorization beyond Devin Desktop (formerly Windsurf). Devin Desktop and CLI are already FedRAMP High Authorized for workloads with ITAR and DoW IL4, IL5, and IL6 requirements. The company says Devin Security Swarm can find and validate vulnerabilities and open remediation pull requests, and that fleets of Devins can upgrade legacy software 5-40x faster than humans alone.

AIPatrick Cheng used ChatGPT to redesign menus, optimize pricing, and improve marketing after acquiring a struggling restaurant in 2021. He reports more than 200% sales growth and a tripling of catering business, and is now launching NextGen, a nonprofit helping family-owned businesses adopt AI through education and micro-grants.

AIJohn Schulman says Thinking Machines, founded a year and a half ago, is revisiting its original instincts after major progress in agents. The company published a written statement of what it now believes about customizable models and humans working well with AI.

AIThe essay argues that AI labs selling model inference face commodity pricing pressure, but may achieve durable profits by moving into products, enterprise deployments, and switching-cost moats. It cites historical infrastructure industries and the Bertrand paradox to support the view that value capture depends on climbing the stack. The authors also warn that successful lock-in could raise enterprise costs and concentrate power, making early interoperability and portability standards important.

AIZed v1.10 now includes proper support for llama.cpp as a model provider, according to Georgi Gerganov. Zed's own announcement credits Hugging Face with helping make the llama.cpp integration possible.

AIMeta Superintelligence Labs has introduced Muse Spark 1.1, a multimodal reasoning model aimed at agentic tasks, with gains in tool use, computer use, coding, and multimodal understanding. It supports a 1 million token context window and is available in Thinking mode in the Meta AI app and on meta.ai, with developers able to access it through a public preview of the Meta Model API.

Why it matters: The post specifies Muse Spark 1.1's agent, coding, and multimodal gains and its Meta Model API preview access, which helps developers judge its fit for their workflows.

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.

AICognition says its SWE-1.7 model, developed from the open-source Kimi K2.7 Code base, performs as well as or better than leading U.S. frontier models on its new trustworthiness evaluation suite. The suite combines 145 politically sensitive questions, sampled in English and Chinese, with realistic coding scenarios to measure propaganda, censorship, and security behavior. Cognition says SWE-1.7 improves substantially over the base Kimi K2.7 Code model, though the company says the benchmarks are still in development.

AIllama.cpp has added DFlash support to its speculative decoding options, joining MTP, Eagle3, and various ngram-based techniques. The author says these methods further improve local model performance. The change was developed with thanks to the NVIDIA team and Ruixiang Wang.

AIAnthropic's Claude Cowork has started rolling out to web and mobile. Chat and Cowork now share one home tab on web and desktop, with a single sidebar, search, and place for Projects and Artifacts. Further Chat and Cowork integration is expected soon.

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.

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.

AIMistral CEO Arthur Mensch says enterprises should use open-source models because closed providers that force data retention gain leverage over their business. He argues companies should store data in open systems, control AI access rules, and build continuous training loops to shrink costs and create hard-to-copy systems. Mistral offers its Studio control plane and Forge training platform, deployed on customer infrastructure or through zero-data-retention hosting.

AIA model repository named XiaomiMiMo/MiMo-V2.5-DFlash is listed on Hugging Face with 311B parameters and tensor types F32, BF16, and F8_E4M3. The README is empty, and the page reports 434 downloads last month and no Inference Provider deployment.

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.

AIWerner Vogels praises Zalando's engineering on client-side load balancing, though he is not a fan of doing it client-side. He especially likes the lessons-learned section that concludes Zalando's write-up.

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.

AITogether AI says it is serving 400 trillion tokens per month, and demand for open models keeps rising. The company also announced an $800 million Series C at an $8.3 billion valuation, as it works to build an efficient generative AI platform.

AIZ.ai has launched ZCode, its official development environment for GLM-5.2, available on macOS, Windows, and Linux. GLM Coding Plan subscribers get 1.5x usage quota in ZCode, and the tool supports bring-your-own-key (BYOK) use with existing subscriptions and APIs.

AIBridgewater partnered with Thinking Machines on Tinker to fine-tune a model that helps its analysts focus on what matters most. The post frames the work as experts improving AI that empowers experts, with a linked article on replicating expert judgment in financial tasks.

AIAndrew Ng describes three loops he uses to build 0-to-1 products with AI agents: an agentic coding loop, a developer feedback loop, and an external feedback loop. He says the agentic coding loop runs every few minutes, letting coding agents build, test, and iterate on software for around an hour without human intervention. The developer feedback loop operates over tens of minutes to hours, with humans steering product decisions because they hold a context advantage over AI systems.

AIAmazon CTO Werner Vogels argues that the "two-pizza" team concept was never about feeding engineers but about ownership, speed, and avoiding bureaucracy. He says working backwards from the customer and writing documents to force clarity remain core practices. He adds that the industry is changing and it is time to reconsider how products are brought to life.

AITri Dao says Etched designed and produced its chips within two years by hardcoding attention into silicon and reaching high MFU. He expects hardware built for LLM inference to cut the cost of intelligence by 10x. The quoted Etched post says it has built its first racks after an A0 tapeout, raised $800m, holds $1B+ in customer contracts, and plans to ship the racks this summer.

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.

AIHamel Husain argues that an AI product whose output is hard to verify is a product design problem, not just an evaluation problem. He shows before-and-after sketches for an AI data agent, a PE lesson planner, and a workers' compensation report tool, each adding provenance, scoped edits, and checkable evidence. He notes that designing for verification also makes evals easier to build and grade.

AIUnlimited-OCR, Baidu's long-context OCR model, now runs in vLLM, with a recipe provided for developers to try it. The background post says it parses entire books in one pass using Reference Sliding Window Attention (R-SWA), which keeps the KV cache fixed during decoding, and claims 35% faster throughput than DeepSeek-OCR at 6K output tokens.

AIPaddleFormers 1.2 is released with support for training DeepSeek-V4 and 128K+ long-context training. The update adds Context Parallel, Packing, Document Mask Attention, and the Muon optimizer, plus ultra-fused mHC, CSA, and HCA operators, DeepEP/HybridEP communication, and lossless FP8 training with AutoSubbatch memory balancing. The project is presented as fully open-source and is available on GitHub.