2026-07-09

OpenAI has launched the GPT-5.6 series, featuring three models: Sol (flagship), Terra (balanced), and Luna (cost-efficient). The flagship Sol model achieves a state-of-the-art score of 7.8% on the ARC-AGI-3 benchmark, becoming the first verified frontier model to beat an ARC-AGI-3 game. This release marks a significant leap in AI reasoning and agentic capabilities, particularly on the challenging ARC-AGI-3 benchmark that measures interactive reasoning and adaptation. The GPT-5.6 series strengthens OpenAI's competitive position against rivals like Anthropic's Fable 5, and the introduction of max/ultra reasoning, multi-agent collaboration, and Programmatic Tool Calling enables more efficient complex task execution. The GPT-5.6 series includes Sol, Terra, and Luna models, with Sol being the most capable. GPT-5.6 Sol sets a new SOTA on ARC-AGI-3 at 7.8% using max reasoning effort, and also improves performance on coding, science, and cybersecurity tasks. The developer guide highlights improved intent understanding and image detail preservation, while GPT-5.4 is scheduled to be deprecated on July 23.

Microsoft has released TypeScript 7.0, a native version rewritten in Go that delivers 8 to 12 times faster build times compared to its predecessor, and supports shared-memory multithreading. Users can install it via npm, and editors supporting LSP can use the new language server. This rewrite represents a paradigm shift in compiler performance for TypeScript, one of the most widely used languages, drastically reducing build times for large projects. It also demonstrates the growing trend of using systems languages like Go to optimize tooling performance. New experimental flags --checkers and --builders allow fine-tuning parallelism for type-checking and project reference building. However, toolchains for embedded languages like Vue and Svelte are not yet ready, requiring those projects to continue using older TypeScript versions.

The European Parliament voted on July 8, 2026, to allow mass scanning of private messages until April 3, 2028, despite a majority of MEPs opposing the measure. This decision represents a significant setback for digital privacy in the EU, as it permits warrantless scanning of private communications on platforms like Gmail, Snapchat, and Skype, setting a precedent for mass surveillance. The motion to reject the scanning law failed because it required an absolute majority of all 705 MEPs (361 votes), not just those present; only 314 voted against, 276 in favor, 17 abstentions, and 113 were absent.

A project called pgrust has rewritten PostgreSQL entirely in Rust using LLMs, and now passes 100% of the official PostgreSQL regression tests. This demonstrates the feasibility of using AI to rewrite large, critical infrastructure software, potentially enabling faster innovation and safer memory management in databases. It also sparks debate on licensing, code review, and trust in AI-generated code. The rewrite uses LLMs to generate code and has produced over 7,000 commits in less than a month, making traditional code review impractical. The project is licensed under AGPL, which differs from PostgreSQL's permissive license.

Ant Group open-sourced LingBot-Video, the world's first mixture-of-experts (MoE) based embodied video foundation model for robotics, achieving a state-of-the-art score of 0.620 on the RBench benchmark. The model is released under Apache 2.0 license on GitHub. This open-source release significantly lowers the barrier for embodied AI research, providing a highly efficient MoE architecture that activates only 3B of 30B total parameters, making it three times more efficient than dense models of similar size. It can accelerate progress in robot action prediction, simulation data generation, and world model development. LingBot-Video innovates in three aspects: architecture (DiT+MoE for capacity-cost balance), data (70K hours of embodied data covering dexterous manipulation, robot locomotion, and egocentric interaction), and training (multi-dimensional reinforcement learning rewards focusing on physical plausibility and task completion). The model uses a Diffusion Transformer (DiT) backbone.

On July 9, 2026, the National Supercomputing Internet core node officially went live in Zhengzhou, offering over 100,000 domestic AI accelerator cards as a pooled computing resource. This milestone marks the largest single domestic AI computing resource pool connected to China's National Supercomputing Internet, significantly boosting the country's sovereign AI compute capabilities and reducing dependence on foreign hardware. The node handles core functions such as operation management, resource scheduling, and integrates supply-demand matching and industry incubation services, aiming to build a nationwide coordinated computing resource system.

Meta publicly launched Muse Spark 1.1 on July 9, 2026, a multimodal AI model designed for agentic coding, with API access available through Meta's developer platform. This release positions Meta as a major competitor to OpenAI and Anthropic in the AI coding assistant space, offering aggressive pricing at $1.25 per million input tokens and potentially disrupting the market by commoditizing coding models. The model is evaluated on Terminal-Bench 2.1, but community analysis highlights that resource limits (6 CPU cores, 8GB RAM) were overridden, which may disqualify results. Pricing is $1.25/$4.5 per million tokens for input/output, with $0.15 for cached input.

Jarred Sumner announced the successful rewrite of the Bun JavaScript runtime from Zig to Rust, a process that took 11 days of intensive agentic engineering using Claude and cost approximately $165,000 in API tokens. This rewrite demonstrates that modern AI coding agents can make large-scale rewrites feasible, challenging the long-held 'never rewrite' wisdom. It also shows how Rust's memory safety can eliminate common bugs like use-after-free that plagued Zig's manual memory management. The Bun test suite, written in TypeScript, served as a conformance suite to validate the port. The new Rust-based Bun has been running in Claude Code since June 17, 2026, with 10% faster startup on Linux and no noticeable changes for users.

OpenAI introduced GPT-Live, a new family of voice models for ChatGPT that enables full-duplex conversation and can delegate complex tasks to GPT-5.5. Two versions are available: GPT-Live-1 for paid users and GPT-Live-1 mini as the default for all users, replacing Advanced Voice Mode. This upgrade significantly enhances ChatGPT voice mode by using a more capable model and seamless task offloading, improving user experience for complex queries. It demonstrates practical progress in conversational AI, making voice assistants more useful for brainstorming and multi-step tasks. GPT-Live uses a full-duplex architecture that can listen and speak simultaneously, maintaining natural conversation flow. It delegates web search, deeper reasoning, and other complex work to OpenAI's latest frontier model GPT-5.5, and OpenAI will continuously update the background model. The larger GPT-Live-1 model is available on paid plans, while GPT-Live-1 mini is free.

Meta's superintelligence unit released a one-year progress update highlighting the most aggressive compute ramp ever seen, including a 2000km+ scale-across infrastructure, and the emergence of a top-tier reinforcement learning environment startup. This signals an acceleration in AI infrastructure competition, with Meta's aggressive investment potentially reshaping the superintelligence landscape. The RL environment startup fills a critical gap in training advanced AI agents. The compute ramp involves 'scale-across' networking spanning over 2000km, enabling multiple data centers to act as a single supercomputer. The RL environment startup is described as emerging 'out of thin air,' indicating a sudden new entrant in the space.

Rust 1.97.0 introduces a new default symbol-mangling scheme, enables denying Cargo warnings via a new flag, and stops hiding linker output after a successful build. These changes improve build reliability and debugging for Rust developers. Denying warnings can help enforce code quality in CI pipelines, while the new mangling scheme ensures more stable symbol names. The new symbol-mangling scheme produces more consistent symbol names across different Rust versions. The Cargo deny warnings feature allows treating warnings as errors, similar to -D warnings in rustc. Removal of hidden linker output means all linker messages are now visible.

IMGNet introduces a face verification model that replaces cosine similarity with sliding window sign pattern matching, achieving 96.27% accuracy on LFW with a 10.58 MB model trained on CASIA-WebFace. This novel approach could enable more efficient and compact face verification systems, and the finding that sign pattern matching improves ArcFace embeddings without retraining suggests a fundamental property of well-trained face embeddings. The model uses a SW Block replacing standard convolution with multi-scale relational operations, an IMG Sign MSE Loss defined purely over sign pattern agreement, and a voting system combining three metrics. Applied to ArcFace (buffalol) without retraining, IMG Sign Score achieves 99.58% on LFW, only 0.24% below ArcFace+Cosine.

DJI's unreleased EV50 VTOL cargo drone flew at 8,861 meters above Mount Everest, setting a world record for the highest altitude achieved by an electric vertical takeoff and landing (eVTOL) drone in public testing. This achievement demonstrates the feasibility of using VTOL cargo drones for high-altitude logistics and scientific research in extreme environments, potentially transforming how supplies are delivered to remote mountainous regions. Over a 12-day mission, the EV50 completed 32 takeoffs and landings, climbed continuously for 3,730 meters, and still had 30% battery remaining on return. It carried ozone-measuring equipment for Peking University researchers.

Meta plans to start mass production of its self-designed AI chip, codenamed 'Iris,' in September 2026, aiming to double its overall AI computing capacity to 14 GW by 2027. This move reduces Meta’s reliance on external suppliers like Nvidia and AMD, potentially reshaping the AI hardware market and lowering costs for large-scale AI training and inference. The chip, part of the MTIA (Meta Training and Inference Accelerator) fourth-generation project, was developed with Broadcom and manufactured by TSMC, passing tests in only six weeks without significant issues.

OpenAI published a set of national security principles that explicitly prohibit the use of its technology for autonomous weapons, mass surveillance, and high-risk automated decision-making. The company also expanded defensive collaborations with US allies under the Daybreak cyber defense program, including partnerships with Australia, Canada, Japan, and EU entities. This sets a clear precedent for AI governance in national security, balancing ethical boundaries with defensive applications. It could influence other AI companies and government policies worldwide, especially regarding the military use of AI. The principles impose a hard ban on mass domestic surveillance, autonomous weapon systems, and high-risk automated decisions. The Daybreak initiative provides advanced AI tools for verified defenders, pairing more permissive features with stronger oversight and scope controls.