ai-powered-markdown-translatorArticle translated from fr to en with gpt-5.4-mini.
June 16, 2026 โ Z.ai releases the MIT open-source weights for GLM-5.2 (1M-token window, two reasoning levels), Alibaba Qwen unveils the Qwen-Robot Suite made up of three foundational models for robotics, NVIDIA Blackwell dominates the full set of MLPerf Training 6.0 benchmarks โ including DeepSeek-V3 671B trained in 2.02 minutes on 8,192 GPUs. GitHub announces paid general availability (general availability) for Code Quality on July 20 at $10 per active developer per month, the June Pixel Drop brings Gemini Omni video and music generation to Pixel devices, and Anthropicโs economic research quantifies for the first time the impact of domain expertise on Claude Code.
Z.ai GLM-5.2 โ available MIT open-source weights
June 16 โ Z.ai (formerly Zhipu AI) officially releases the GLM-5.2 open-source weights under the MIT license, making good on the June 13 announcement. The weights are now available for download on the companyโs platform.
- Significant improvements in coding and agentic tasks
- Strong long-horizon capabilities with a 1M context window
- Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balanceโ โ @Zai_org on X
Key points of GLM-5.2:
- Significant improvements on coding and agentic tasks
- 1 million token context window โ useful for large-scale coding agents and processing long documents
- Two complementary levels of reasoning effort
- Outperforms GLM-5.1 on coding, tool use, and reasoning
- Available via chat.z.ai
| Variant | Reasoning | Context | License |
|---|---|---|---|
| GLM-5.2 (max) | Maximum performance | 1M tokens | MIT open-source |
| GLM-5.2 (high) | Strong performance/speed balance | 1M tokens | MIT open-source |
The MIT license is the most important marker: it allows unrestricted commercial use, making GLM-5.2 directly usable by companies that cannot or do not want to depend on proprietary U.S. models. The tweet reached 515,000 views within hours โ a strong signal of the communityโs interest in frontier open-source models with large context windows.
Qwen-Robot Suite โ three foundational models for embodied robotics
June 16 โ Alibaba Qwen announces the Qwen-Robot Suite, a complete stack for embodied intelligence made up of three distinct and complementary foundational models: navigation, manipulation, and world model.
Qwen-RobotNav โ specialist in mobile navigation:
- Unifies 5 tasks into a single model: instruction following, point-goal navigation, object-goal navigation, object tracking, and autonomous driving
- Controllable observation protocol
- Tool interface for agentic systems
Qwen-RobotManip โ specialist in manipulation:
- Unified state-action space for heterogeneous robots
- Camera-frame delta poses for consistent learning across different robot bodies (cross-embodiment)
- Pre-trained on an open-source corpus of more than 38,100 hours
Qwen-RobotWorld โ physical world model:
- One model covering more than 20 robot body types
- Natural-language action interface
- Predicts physically grounded futures for manipulation, driving, and navigation
Each model can be used independently and can be combined as a tool of the physical world. Together, they form the low-level toolkit for general-purpose agentic systems.
| Model | Specialty | Key data |
|---|---|---|
| Qwen-RobotNav | Multi-task navigation (5 unified tasks) | Tool interface for agentic systems |
| Qwen-RobotManip | Cross-embodiment robotic manipulation | 38,100+ hour open-source corpus |
| Qwen-RobotWorld | Physical world model | 20+ robot body types |
๐ Qwen-Robot Suite โ announcement
NVIDIA Blackwell dominates MLPerf Training 6.0 โ clean sweep
June 16 โ NVIDIA won every MLPerf Training 6.0 benchmark, the industry standard for AI training established by the MLCommons consortium. It is the only platform to have submitted results on every test, including the two new benchmarks introduced in this edition: DeepSeek-V3 (671 billion parameters, MoE architecture) and GPT-OSS-20B.
The performance of the GB300 NVL72 system (72 interconnected Blackwell Ultra GPUs) is remarkable:
| Model | Platform | GPU | Time |
|---|---|---|---|
| DeepSeek-V3 671B (MoE) | GB300 NVL72 | 8,192 | 2.02 min |
| GPT-OSS 20B (MoE) | GB300 NVL72 | 512 | 7.43 min |
| Llama 3.1 405B | GB200 NVL72 | 8,192 | 7.07 min |
| Llama 3.1 8B | GB200 NVL72 | 1,024 | 4.46 min |
| Llama 2 70B LoRA | GB300 NVL72 | 512 | 0.40 min |
| FLUX.1 (image generation) | GB300 NVL72 | 512 | 17.1 min |
| DLRM-dcnv2 | GB300 NVL72 | 64 | 0.67 min |
The inclusion of FLUX.1 (Black Forest Labs image generation model) in the official benchmarks is notable: it confirms that the Blackwell infrastructure is the reference standard for training the most advanced image-generation models.
On the software side, NVIDIA rolled out several innovations in the NeMo 26.06 stack: full-iteration CUDA graphs for MoEs, kernel fusions via CuTe DSL, MXFP8 precision for the attention block, and MoE router optimizations (5ร kernel acceleration). These software improvements delivered a 1.3ร increase in DeepSeek-V3 throughput over three months without any hardware changes, from 1,298 to 1,648 TFLOPS/GPU โ a reminder that hardware performance is only part of the equation.
๐ NVIDIA โ MLPerf Training 6.0
June Pixel Drop 2026 โ Gemini Omni video, music generation, Daily Brief
June 16 โ The June Pixel Drop 2026 introduces several major features that bring Gemini directly into Pixel devices, with the gradual rollout beginning in the coming weeks.
Gemini Omni โ AI video creation and editing: just talk naturally with Gemini to mix text, images, and videos. You can start from scratch, remix content from the gallery, use preset templates, or even create a custom AI avatar in the userโs image and voice.
Music generation in Gemini: the user describes an idea or imports a photo, which is turned into an original audio track with lyrics. Style, voice, and tempo can be customized from the Tools โ โCreate musicโ menu in the Gemini app.
Daily Brief โ AI morning agent: available to all Google AI Plus, Pro, and Ultra subscribers (18+, U.S. only for now). This agent organizes the userโs day before they even wake up. Voice Translate, which translates calls in real time, is expanding to the Pixel 10a (7 languages, including French).
| Feature | Availability | Notes |
|---|---|---|
| Gemini Omni video | Pixel (gradual rollout) | Conversation-based creation/remix |
| Music generation | Pixel (Gemini app) | Audio + lyrics, customizable |
| Daily Brief | Google AI Plus/Pro/Ultra, 18+, U.S. | AI morning agent |
| Voice Translate | Pixel 10a | 7 languages + Hindi in preview |
| Edit with Ask Photos | UK, Germany, France, Spain, Italy | Conversation-based photo editing |
๐ June Pixel Drop 2026 โ official Google blog
GitHub Code Quality โ general availability on July 20, $10 per active developer
June 16 โ GitHub announces that GitHub Code Quality will move from public preview to general availability (general availability) on July 20, 2026, as a paid product. More than 10,000 companies participated in the preview.
Pricing on July 20:
| Component | Pricing |
|---|---|
| License per active developer (committer) | $10 / active developer / month |
| AI features (Copilot code review, AI detection, Copilot Autofix) | Usage-based billing |
| CodeQL deterministic analysis | GitHub Actions minutes |
New features available at GA:
- Organization-scale rollout (single admin toggle)
- Organization-level quality dashboards
- Enforcement of code coverage through rulesets
- Quality score by repository and by organization
- API for enabling and managing outcomes
Availability: GitHub Enterprise Cloud and GitHub Team only. Not available on GitHub Enterprise Server. Customers who want to opt out before billing can disable Code Quality on their repositories before July 20.
๐ GitHub Code Quality โ GA announcement
Anthropic economic research โ domain expertise outweighs coding skill
June 16 โ Anthropic publishes an economic study on Claude Code usage, based on confidential analysis of about 400,000 sessions between October 2025 and April 2026. The first quantified results on how expertise determines success with an AI coding agent.
Key findings from the study (authors: Zoe Hitzig, Maxim Massenkoff, Eva Lyubich, Ryan Heller, Peter McCrory):
| Indicator | Value |
|---|---|
| Sessions analyzed | ~400,000 |
| Distinct users | ~235,000 |
| Coverage period | Oct 2025 โ Apr 2026 |
| Planning decisions (user) | ~70% |
| Execution decisions (Claude) | ~80% |
| Average economic value uplift of tasks | +25% |
| Verified success rate (domain expert) | 28โ33% |
| Verified success rate (novice) | 15% |
| Share of debugging sessions Oct โ Apr | 33% โ 19% |
Core conclusion: in a typical session, the user makes ~70% of planning decisions (what to do) while Claude makes ~80% of execution decisions (how to do it). Domain experts โ not necessarily developers โ steer Claude toward action chains that are twice as long and contain five times more content per instruction. The success rate of non-technical professions (lawyers, managers, scientists) is within less than 7 percentage points of software engineers. It is understanding the business problem, not coding skill, that determines success.
๐ Anthropic economic research โ Claude Code
Anthropic and OpenAI โ production agents and European rollout
Claude Managed Agents โ production guide
June 16 โ Anthropicโs Applied AI team publishes a practical guide on deploying agents with Claude Managed Agents. The post explains the evolution of interaction surfaces: from the classic chatbot to planned, persistent autonomous agents with access to execution environments (stateful runtime environments). Two features available since June 9: running agents on a defined schedule and storing environment variables in vaults (vaults).
๐ Building with Claude Managed Agents
Claude Code v2.1.178 โ granular permissions and nested skills
June 15 โ Claude Code v2.1.178 introduces the Tool(param:value) syntax for permission rules โ for example Agent(model:opus) to block Opus sub-agents. Skills in nested .claude/skills/ directories now load automatically when working in those folders. Auto mode now evaluates sub-agent launches through the classifier before startup, closing a loophole where a sub-agent could request a blocked action without prior review.
๐ Claude Code v2.1.178
Codex app in EEA, UK and Switzerland โ Computer Use, Memories, Chronicle
June 16 โ OpenAI rolls out four advanced Codex features in the European Economic Area (EEA), the United Kingdom, and Switzerland: Computer Use (macOS + Windows), the Codex Chrome extension, Memories, and Chronicle.
Computer Use lets Codex interact with desktop apps โ see, click, and type text. The Chrome extension makes it possible to carry out tasks that require connected browser context, working across multiple tabs in the background. Memories stores recurring repository preferences and conventions, but is disabled by default in the three regions for regulatory compliance. Chronicle, an opt-in preview reserved for Pro macOS subscribers, helps Codex build memories from the recent screen context. These features were already available to Enterprise users outside the EEA/UK/CH.
๐ OpenAI โ Codex rollout EEA/UK/Switzerland
OpenAI Deployment Simulation โ predicting behavior before release
June 16 โ OpenAI publishes its Deployment Simulation method: replaying real anonymized conversations from previous deployments with a candidate model to detect undesirable behaviors before production rollout. Tested on GPT-5-series Thinking deployments (~1.3 million conversations), the median prediction error is 1.5ร, significantly better than synthetic evaluations. The method made it possible to detect โcalculator hackingโ (a form of reward hacking) before the release of GPT-5.1.
๐ Deployment Simulation โ OpenAI
Codex CLI 0.140.0 โ /usage, /import from Claude Code, Bedrock auth
June 15 โ Version 0.140.0 of Codex CLI brings several substantial features. /usage displays the accountโs daily, weekly, and cumulative token activity. /import lets you selectively import configuration, project settings, and recent discussions from Claude Code โ direct interoperability between two competing tools. codex delete permanently deletes a session with safety confirmation and cleanup of associated subagents. Managed Amazon Bedrock authentication stores API keys locally in encrypted form, and also covers MCP OAuth credentials. In parallel, MCP reliability fixes (retries on transient startups) and automatic recovery of corrupted SQLite databases.
๐ Codex CLI 0.140.0 โ GitHub release
Meta AI โ AI Mode on Facebook with Muse Spark
June 16 โ Meta is rolling out new artificial intelligence tools on Facebook. AI Mode is a new search mode powered by Meta AI (driven by Muse Spark) that grounds its answers in what users publicly say across Meta apps โ Groups, Reels โ offering contextualized insights rather than generic results. Accessible from the news feed or search.
New creative tools: sharing suggestions from the camera roll with cutout templates, video transition effects, photo presets that let you virtually change clothing and hairstyle, and the โWear Itโ option to virtually wear a sports jersey in Stories. All these features are opt-in and can be disabled.
๐ Meta โ new AI tools on Facebook
NVIDIA SpatialClaw โ training-free spatial reasoning agent
June 16 โ NVIDIA Research releases SpatialClaw, a training-free visual spatial reasoning agent. Rather than calling a fixed set of predefined tools, the agent writes Python directly in a persistent kernel, dynamically composing perception modules and revising its strategy as it moves through steps. Outputs become reusable Python variables with NumPy and SciPy.
Results: +11.2 points on 20 visual spatial reasoning benchmarks compared with a recent prior agent, consistently across 6 different model architectures, without any benchmark-specific tuning.
๐ SpatialClaw โ @NVIDIAAI on X
Cohere โ surge of demand after US restriction on Anthropic
June 15 โ Cohere, a Canadian generative AI company, reports receiving a high volume of inbound requests from companies looking for alternatives following the US directive restricting access to Anthropic (covered on June 13). The company responded humorously by quoting the Polymarket tweet: โThe world needs more Canada ๐จ๐ฆโ. The tweet reached 56,000 views and 797 likes. This positioning confirms that the US directive is creating a direct business opportunity for non-American sovereign providers.
๐ Cohere โ tweet
GitHub Models removed for new customers
June 16 โ GitHub announces the phased retirement of GitHub Models. Starting June 16, new customers (organizations and enterprises without prior usage) no longer have access to the service, on either free or paid plans. Existing customers with active usage are not affected for now and can continue using the playground, the API, and the available models. GitHub says it will share detailed timelines for the full shutdown later. For new projects needing access to AI models, GitHub points to Azure AI Foundry, which offers an extensive model catalog.
Briefs
-
Claude Code v2.1.179 โ 9 stability fixes (June 16): mid-stream connection drops preserved, mouse-wheel scrolling fixed in WSL2 under Windows Terminal and VS Code (regression since v2.1.172), Linux sandbox fix on large directory trees. ๐ v2.1.179
-
Gemini Trusted Tester Program โ registration open: Google is opening a limited number of spots for power users who want to test unreleased Gemini features before public launch. ๐ Sign up
-
Gemini โ multilingual microphone 70+ languages (Android and iOS): the microphone icon now supports more than 70 languages with free mixing without changing settings. Announced by @joshwoodward (Director of Product, Gemini App).
-
GitHub Code Quality โ one-click organization activation (June 16): organization admins can enable or disable Code Quality across all their repositories with a single toggle in security settings. ๐ Changelog
-
Copilot usage metrics โ enriched by server-side telemetry (June 15): Copilot Enterprise usage reports now include server-detected active users (in addition to client telemetry), improving DAU coverage in daily and 28-day reports. ๐ Changelog
-
Manus โ message queue during execution (June 16): Manus now lets you queue messages to send to the agent while a task is running. Messages are processed in order once the current task finishes. ๐ Announcement
-
ChatGPT iOS 1.2026.160 (June 15): workspace file explorer, folder picker, diff controls, MCP approval via chat or globally, and LaTeX rendering in Codex messages and plans.
What this means
The open MIT weights of GLM-5.2 and the three Qwen-Robot Suite robotic models illustrate the acceleration of the Chinese open-source offer in the segments where American players remain largely proprietary. GLM-5.2, with its 1M-token window under MIT license, positions itself directly as a competitor to commercial models for companies seeking a sovereign alternative โ exactly the niche that the US directive restricting Anthropic made strategically visible, as Cohereโs reported influx of requests confirms. Robotics is the next battleground in this competition: Qwen-Robot Suite offers a complete navigation-manipulation-world model stack, while NVIDIA Researchโs SpatialClaw shows that advanced spatial reasoning can be achieved without fine-tuning โ two complementary approaches that sketch the architecture of the next generation of agentic robotic systems.
On the hardware front, the MLPerf Training 6.0 results confirm Blackwellโs dominance at industrial scale: training DeepSeek-V3 (671 billion parameters) in 2 minutes on 8,192 GPUs is not an academic feat, itโs a real production capability deployed in cloud datacenters. The 1.3ร software gain in DeepSeek-V3 throughput over three months without any hardware change is a reminder that the race for AI performance is as much a competition of software stacks (NeMo, CuTe DSL, MXFP8) as of chips โ a dynamic that amplifies already dominant positions.
Anthropicโs economic research on Claude Code provides an important counterintuitive insight: domain expertise (law, management, science) determines success with a coding agent almost as much as coding skill itself. Verified success rates show only a 7-point gap between engineers and non-developers who are experts in their field. This is a structural signal for product teams: the target audience for AI coding tools is no longer only developers, and interface design must take that into account.
On the developer tooling side, June 16 brings several converging signals: the paid GA of GitHub Code Quality ($10/active developer), the availability of Codex with Computer Use in the EEA/UK/Switzerland, the granular permissions in Claude Code v2.1.178, and OpenAIโs Deployment Simulation method. Code quality and deployment security are moving from research topics to billable products โ a normalization that reflects the growing maturity of the AI tools market for development teams.
Sources
- Z.ai GLM-5.2 โ @Zai_org on X
- Z.ai GLM-5.2 โ official blog
- Qwen-Robot Suite โ @Alibaba_Qwen on X
- Qwen-Robot Suite โ official blog
- NVIDIA Blackwell MLPerf Training 6.0
- June Pixel Drop 2026 โ Google official blog
- June Pixel Drop โ @GeminiApp on X
- GitHub Code Quality GA โ changelog
- Anthropic economic research โ Claude Code
- Anthropic โ @AnthropicAI on X
- Claude Managed Agents โ blog
- Claude Code v2.1.178
- Claude Code v2.1.179
- Codex app EEA/UK/Switzerland โ @OpenAIDevs on X
- Deployment Simulation โ OpenAI
- Codex CLI 0.140.0 โ GitHub release
- Meta AI โ new Facebook tools
- Meta AI โ @MetaNewsroom on X
- NVIDIA SpatialClaw โ @NVIDIAAI on X
- Cohere โ @cohere on X
- GitHub Models retirement
- GitHub Code Quality โ organization activation
- Copilot usage metrics โ changelog
- Manus โ message queue
- Gemini Trusted Tester Program