Apr 10, 2026
Product
HeyGen launches Avatar V — new avatar model with stronger realism from short source footage
HeyGen released Avatar V, a new avatar generation model that produces higher-fidelity avatars from shorter video clips and single images. The model uses improved neural rendering and motion synthesis to reduce the data requirements for avatar creation whilst maintaining realism across facial expressions and gestures.
Why it matters: Lower barrier to entry for avatar creation expands use cases in video localization, customer service automation, and synthetic media production.
Read more on HeyGen →
Apr 10, 2026
Product
Factory releases a desktop app — native macOS and Windows interface for AI agents
Factory, a platform for building autonomous AI agents, released native desktop applications for macOS and Windows. The desktop app provides a native interface for designing, testing, and monitoring agent workflows, reducing latency and improving the development experience compared to browser-based access.
Why it matters: Desktop-first tools for agent development signal a shift towards treating AI automation as a core workflow — not just an experimental chat feature.
Read more on Factory →
Apr 9, 2026
Product
Anthropic launches Project Glasswing — security initiative using Claude Mythos Preview to secure critical software
Anthropic announced Project Glasswing, a security initiative that harnesses Claude Mythos Preview to analyse and improve the security posture of critical software systems. The programme pairs security researchers and engineering teams with an advanced AI assistant trained to identify vulnerabilities, suggest remediation strategies, and audit complex codebases for common security flaws.
Why it matters: Using frontier AI to secure critical infrastructure represents a net positive public-good investment — and positions Anthropic's models as essential to defending against emerging threat vectors.
Read more on Anthropic →
Apr 9, 2026
Product
Google launches Gemma 4 — newest open model family for advanced reasoning and agentic workflows
Google released Gemma 4, the latest model in its open-source Gemma family, with significant improvements in reasoning, instruction-following, and multi-step planning. Gemma 4 is optimised for agentic patterns and can run efficiently on consumer hardware whilst competing with larger proprietary models on reasoning benchmarks.
Why it matters: Open models with frontier-grade reasoning capabilities accelerate adoption in cost-sensitive deployments and competitive pressure on closed-model pricing strategies.
Read more on Google Blog →
Apr 9, 2026
Product
Google Vids adds AI video and music tools — powered by Veo 3.1 and Lyria 3
Google added new AI video and music generation features to Google Vids, powered by Veo 3.1 and Lyria 3 models. Users can now generate video scenes and background music directly within the editing interface, streamlining the workflow for creating marketing and promotional content.
Why it matters: Integrated video and music generation in accessible editing tools democratises professional-grade content production for SMBs without specialist technical skills.
Read more on Google Blog →
Apr 9, 2026
Infrastructure
Anthropic reportedly exploring its own chip design — compute competition intensifies
Sources told Reuters that Anthropic is evaluating options to design and manufacture its own custom AI chips, joining a broader shift towards vertical integration in the AI industry. The exploration reflects concerns about compute availability and pricing as competition for GPU capacity intensifies across all frontier labs.
Why it matters: If Anthropic commits to custom chips, it signals a long-term capital investment in independence from NVIDIA — and validates the strategic importance of owning the entire compute stack.
Read more on Reuters →
Apr 9, 2026
Product
Seedance 2.0 arrives on Runway — text-to-video with image and audio support
Runway released Seedance 2.0, a third-party video generation model now available on the Runway platform, with new capabilities for multi-modal input. The model supports text, image, video, and audio prompts, enabling users to generate multi-shot videos from mixed-media references in a single workflow.
Why it matters: Multi-modal video generation reduces the friction for creators moving from concept to finished footage, opening new use cases in advertising and short-form content.
Read more on Runway →
Apr 9, 2026
Industry
Kia plans to use Boston Dynamics Atlas humanoid robots in US factory from 2029
Kia announced plans to deploy Boston Dynamics Atlas humanoid robots in its Georgia manufacturing facility starting in 2029, as part of a broader strategy to increase automation and adapt to evolving workforce dynamics. The robots will handle repetitive assembly tasks in the EV production line.
Why it matters: Major automakers betting on humanoid robots for assembly signals confidence in the near-term viability and ROI of humanoid platforms for factory deployment.
Read more on Reuters →
Apr 8, 2026
Product
Meta releases Muse Spark — first model from Meta Superintelligence Labs, rolling out via Meta AI app
Meta released Muse Spark, the first production model from its newly established Superintelligence Labs team, with initial availability through the Meta AI app and mobile. The model is tuned for multi-turn conversations and reasoning tasks, with rollout planned across Meta's ecosystem including WhatsApp and Instagram.
Why it matters: Meta consolidating its AI research under a dedicated superintelligence group signals a competitive reset after months of criticism over falling behind in frontier AI capability.
Read more on The Verge →
Apr 8, 2026
Product
Gemini adds interactive 3D models and charts — generates interactive simulations in-app
Google rolled out the ability for Gemini to generate interactive 3D models and dynamic charts directly within the chat interface. Users can now rotate models, adjust parameters, and explore data visualisations without leaving Gemini, making it easier to reason about spatial and quantitative concepts conversationally.
Why it matters: Richer in-app outputs reduce friction for users who want to go from question to working visualisation — improving Gemini's appeal for education and technical work.
Read more on Google Blog →
Apr 8, 2026
Product
Gemini adds notebooks for project organisation — files, context, and syncing to NotebookLM
Google introduced Notebooks as a new Gemini feature, allowing users to organise files, chats, and project context in a single workspace. Notebooks sync bidirectionally with NotebookLM, Google's research tool, enabling seamless collaboration between conversational and long-form analysis workflows.
Why it matters: Linking Gemini's chat interface to structured project spaces positions Google to capture more of the "thinking workflow" lifecycle, competing with specialised research tools.
Read more on Google Blog →
Apr 8, 2026
Product
Claude Managed Agents enters public beta — hosted agentic inference on Claude
Anthropic announced that Claude Managed Agents has entered public beta, offering a hosted service for building and running tool-using agents powered by Claude. The platform handles context management, tool invocation, retry logic, and error handling, simplifying the operational complexity of deploying autonomous agents.
Why it matters: Managed inference for agents lowers the barrier to production deployments, turning agency from an advanced use case into a standard product offering.
Read more on Anthropic →
Apr 8, 2026
Product
OpenAI adds $100 Pro tier for Codex — higher rate limits and faster inference
OpenAI introduced a new $100-per-month Pro tier for Codex, its code-generation API, offering significantly higher rate limits and priority access compared to the Plus plan. The tier is aimed at teams and enterprises running production code-generation workloads that require dedicated capacity.
Why it matters: Tiered pricing for coding APIs signals that enterprise demand for high-throughput code generation justifies premium pricing in the SaaS hierarchy.
Read more on OpenAI →
Apr 7, 2026
Product
Z.ai releases GLM-5.1 — flagship model for agent-style work and advanced reasoning
Z.ai released GLM-5.1, a new flagship model designed for long-horizon agentic workflows and engineering tasks. The model is positioned for complex reasoning chains, coding generation, and autonomous problem-solving with stronger performance on multi-step reasoning compared to its predecessor.
Why it matters: More purpose-built agent models increase competitive pressure on foundational labs to specialise beyond raw capability benchmarks.
Read more on Z.ai →
Apr 7, 2026
Infrastructure
Broadcom signs long-term Google AI chip deal — cementing custom silicon partnerships
Broadcom announced a long-term strategic partnership with Google to develop and supply custom AI chips, including Google's proprietary silicon architectures. The deal extends across data-centre infrastructure and reflects the broader industry shift toward vertically integrated compute, with Anthropic embedded in the wider context of competing chip investments.
Why it matters: Custom silicon deals lock in manufacturing capacity and differentiation — companies investing early in proprietary chips gain lasting cost and performance edges.
Read more on Reuters →
Apr 7, 2026
Industry
Perplexity expands Plaid integration — connecting more account types for AI-assisted finance
Perplexity rolled out deeper Plaid integration, enabling its AI assistant to connect to more account types and financial institutions. The expanded partnership allows users to ask the assistant natural-language questions about their bank balances, transactions, and financial portfolios without leaving the chat interface.
Why it matters: AI assistants gaining direct financial-data access represent a shift in how consumers interact with their money — and where financial data flows in the tech stack.
Read more on Plaid →
Apr 7, 2026
Industry
Skild AI deploys robotics software on Foxconn assembly lines — running inference on Nvidia Blackwell systems
Skild AI announced the deployment of its robotics software stack on Foxconn manufacturing lines, where it is running inference workloads on Nvidia Blackwell chips. The partnership puts visual and control models in direct production use, handling real-time decision-making for assembly tasks building high-end AI hardware.
Why it matters: Robotics models in production manufacturing prove the economics and reliability of AI-powered automation at scale — and feed data back into the next iteration of models.
Read more on Reuters →
Apr 7, 2026
Industry
Agile Robots partners with Google DeepMind — integrating Gemini Robotics models into production systems
Agile Robots announced a partnership with Google DeepMind to integrate the Gemini Robotics model family into its manipulation and control systems. The collaboration pairs Agile Robots' specialised hardware with Google's multi-modal perception and reasoning models, targeting industrial automation workflows.
Why it matters: Major robotics firms integrating foundation models from tech giants signals that custom robotics models are converging on commodity infrastructure.
Read more on TechCrunch →