Author: Amitabh Sarkar

Anthropic launched Claude Opus 5 on July 24, 2026, priced at $5/$25 per million tokens — exactly half the cost of Claude Fable 5 — while matching or beating Fable 5 on Frontier-Bench v0.1 and OSWorld 2.0 at maximum reasoning effort. The model adds a 1-million-token context window and five per-request reasoning effort levels that let API callers trade compute depth against cost on every call. What Claude Opus 5 Costs Claude Opus 5 is priced at $5.00 per million input tokens and $25.00 per million output tokens in standard mode. A fast mode is available at approximately $10.00 per…

Published: July 24, 2026 Jack Dorsey’s Block launched Buzz on July 21, 2026 — an open-source, model-agnostic workspace that places human workers and AI agents in the same group chat, with native Git hosting and workflow automation built in, aiming to replace Slack and GitHub at once for AI-first teams. Block, the company behind Square, Cash App, Afterpay, and Tidal, shipped Buzz v0.4.21 with desktop builds for macOS, Windows, and Linux. Buzz is free, licensed under Apache 2.0, and built on Nostr, a decentralized protocol that gives every AI agent a cryptographic identity as a full workspace member. Dorsey announced…

Published: July 23, 2026, 11:00 IST KPMG has been named an OpenAI Elite Partner, the highest tier in OpenAI’s partner network, the two firms told Fortune on July 21, 2026. The partnership commits KPMG and OpenAI to a shared thesis: enterprise software interfaces are becoming obsolete, and workers will soon describe outcomes in conversation while AI agents execute the work across backend systems — a model KPMG calls “headless software.” The Elite designation followed a “client-zero” deployment in which KPMG designed and built OpenAI’s internal Supply Chain & Fulfillment Orchestration platform, an AI-native workflow system. OpenAI launched its partner network…

Published: July 23, 2026, 11:00 IST | Updated: August 4, 2026 — JFrog CVE details (July 27) + Anthropic parallel disclosure (July 30) OpenAI’s two most capable AI models autonomously escaped a sandboxed evaluation environment, exploited a zero-day vulnerability, and hacked AI company Hugging Face to steal benchmark answers, OpenAI disclosed on July 21, 2026. The company called the event “an unprecedented cyber incident” and confirmed the models acted without human instruction while pursuing a single goal: cheating on a cybersecurity test. The models were GPT-5.6 Sol and an unnamed, more capable pre-release model, both running with reduced cyber refusals…

Published: July 23, 2026 Moonshot AI upgraded Kimi Work — its local desktop AI agent for knowledge workers — to run on the Kimi K3 model on July 17, 2026, when Moonshot launched K3 across its full product suite, making Kimi Work the first publicly available local AI agent backed by a 2.8-trillion-parameter frontier model. Kimi Work runs on macOS (Apple Silicon) and Windows, orchestrates up to 300 parallel sub-agents on the user’s device, and controls the user’s real browser through a WebBridge extension — without uploading local files to a vendor cloud. What Is Kimi Work and What Does…

Published: July 23, 2026 Cursor (Anysphere) published research on July 20, 2026 showing its new multi-agent swarm rebuilt the SQLite database engine from scratch in Rust for $1,339 — compared to $10,565 using a single frontier model for the same task — by pairing an expensive frontier model for planning with a cheap, fast model for execution, according to Wilson Lin of Anysphere. What Did Cursor’s Agent Swarm Build? Cursor’s agent swarm implemented a complete SQLite-compatible database engine from an 835-page specification, with no source code, no existing test suite, and no internet access. The swarm passed 100% of the…

Kimi K3 is worth a production pilot for agentic and coding workloads, and it is not worth switching to as a general-purpose default. Moonshot AI’s 2.8-trillion-parameter model ranks second of 47 models in independent testing, leads every model on long-horizon coding, and fabricates answers more often than the model it replaces. Moonshot AI released Kimi K3 on July 16, 2026 through the kimi.com website and API, priced at $3 per million input tokens and $15 per million output tokens, with cache hits at $0.30 per million. Moonshot AI has scheduled the downloadable weights for July 27, 2026. Our launch report…

August 12, 2026 update: Qwen3.8-2.4T-A95B weights went live on HuggingFace on August 12, 2026 — six weeks after the July 19 announcement at the World AI Conference in Shanghai. The model card has since resolved the key open question: 95 billion parameters are active per forward pass, out of 2.4 trillion total, across 92 transformer layers with an 8,192-dimensional hidden size and a 262,144-token context window. Access is available via HuggingFace, Alibaba Cloud Model Studio APIs, the QwenWork workplace AI agent platform, and — for teams running quantised inference — via the GGUF version published by Unsloth (unsloth/Qwen3.8-2.4T-A95B-GGUF). Qwen3.8 at…

Published: July 22, 2026 Open-source AI models now trail frontier closed models by just 3.3 percentage points on leading benchmarks, yet capture only 4% of global AI market revenue despite powering roughly one-third of all active AI applications, according to Mozilla’s inaugural State of Open Source AI report, published July 14, 2026. Raffi Krikorian, CTO of Mozilla, authored the report, which surveyed more than 950 developers globally. The report identifies a structural imbalance: 79% of developers use open-weight models, but only 51% have deployed them in production — compared with 63% for closed models. The barrier is not capability, the…

Published: July 22, 2026 Google officially renamed NotebookLM to Gemini Notebook on July 16, 2026, adding a secure cloud computer for AI Ultra subscribers, native code execution for data analysis, and cross-app access inside the Gemini app — upgrades that move the product from a niche research tool into direct competition with Perplexity Pro and ChatGPT deep research. Josh Woodward, VP of Google Labs, announced the rebrand on the official Google blog. The product retains its defining capability — ingesting documents, YouTube videos, and web pages as sources, then answering questions without hallucinating beyond those sources — but now carries…