Track the labs, vendors, coding tools, and physical AI companies shaping the market right now.

Linking Qwen releases with cloud adoption, enterprise tooling, and China-scale product reach.

Anthropic's current lineup centers on Claude Opus 4.6 and Sonnet 4.6 for coding, agents, and long-context enterprise work.

CoreWeave is leaning further into its public-market AI cloud story, pairing infrastructure scale with new storage, capacity, and platform options for training and inference workloads.

DeepSeek's current public stack is led by DeepSeek-V3.2 for agentic tool use, V3.2-Speciale for maximum reasoning, and the updated R1-0528 reasoner.

Google DeepMind's current flagship model set is led by Gemini 3.1 Pro, with newer 3.1 Flash variants and an active robotics track.

Hugging Face continues to act as both the model commons and a builder of lightweight open families like SmolLM3 and SmolVLM2.

Continuing to use open-weight releases as a distribution wedge across Meta products.

Microsoft continues to push Azure and Copilot distribution while extending Phi with reasoning and multimodal variants like Phi-4-reasoning-vision.

MiniMax is positioning itself as a global multimodal model and AI-product company with very large user reach.

Pushing Kimi toward broader reasoning and productivity use cases while defending strong domestic consumer adoption.

Turning chips, simulation, and model families into a full-stack platform for physical and enterprise AI.

OpenAI's current flagship API model is GPT-5.4, with GPT-5 mini for lower-latency workloads and GPT-5.3-Codex for agentic coding workflows.

Perplexity is now pushing beyond basic answer search into Pro Search, Deep Research, and API access built around its Sonar model family.

Alibaba Cloud has moved Qwen forward again with the Qwen3.5 series, pairing open-weight releases with hosted Plus and Max-Thinking variants.

Shield AI keeps expanding Hivemind and V-BAT as autonomy products for increasingly contested defense environments.

Thinking Machines Lab is scaling up frontier training and platform ambitions through its NVIDIA partnership and Tinker research platform.

Waymo is continuing to expand paid autonomous ride-hailing across U.S. cities while extending its international footprint.

xAI has deployed Grok 4.1 and Grok 4.1 Fast, while also expanding into media generation with the Grok Imagine API.

Z.ai's current releases now include GLM-5, newer GLM-4.7 variants, and an actively maintained image stack led by CogView-4.

1X is pushing toward home deployment with NEO and Redwood AI while consolidating operations around its Palo Alto headquarters.

Abridge is expanding its contextual reasoning and clinical documentation platform across major health systems.

AI2's latest open-model push now highlights OLMo Hybrid, alongside Molmo and Tulu, as part of its open research stack.

AWS now positions Amazon Nova as a broader platform that includes Nova 2 Pro, Nova 2 Lite, Nova 2 Sonic, and Nova Act alongside creative models like Nova Reel.

Anduril continues scaling Lattice and autonomous defense systems across manufacturing and government programs.

Anyscale continues to position itself as the production-ready platform for teams building on Ray.

Augment is expanding from IDE assistance into broader agent, CLI, and code-review workflows for professional engineering teams.

Baseten is continuing to position itself as a fast, reliable inference layer for open-source and custom models.

Black Forest Labs is now centering its public lineup on FLUX.2 [klein], FLUX.2 [max], and FLUX.2 across API, self-hosted, and playground distribution.

Boston Dynamics is expanding commercial deployments of Spot and Stretch while pairing its robotics stack with newer AI and autonomy partnerships.

ByteDance continues expanding Doubao and multimodal media tooling as a consumer AI layer across its broader ecosystem.

Cambricon continues to position its domestic AI processors as part of China's push for more self-sufficient AI compute infrastructure.

Canva keeps embedding Magic Studio features deeper into design, docs, and presentation workflows.

Cerebras is pushing its wafer-scale stack deeper into production inference through partner deployments and low-latency serving offers.

Claude Code is becoming one of Anthropic's clearest developer-facing wedges, centered on terminal workflows, repo-aware editing, and automated engineering tasks.

Clay is broadening from enrichment into agentic outbound and GTM workflow automation.

Codex is now a distinct OpenAI developer product aimed at turning model capability into hands-on coding, planning, and execution workflows for software teams.

Cohere's current lineup is led by Command A, with newer enterprise multimodal and reasoning variants for vision and agent-heavy workflows.

Credo AI is expanding from governance messaging into model and vendor evaluation workflows, including Model Trust Score and registry-style enterprise tooling.

Crusoe is pushing its renewable-powered AI factory positioning as demand for large-scale training and inference infrastructure grows.

Cursor continues to push deeper agentic software workflows, combining IDE-native autocomplete with repo-aware editing and background work.

d-Matrix is pushing Corsair and JetStream as efficient inference systems for batched generative AI workloads.

Blending model releases with enterprise data platform lock-in.

DataRobot is now positioning itself as a unified agent workforce platform for enterprise AI.

DeepL continues expanding from translation into writing and voice products for enterprise and professional users.

Extending from voice generation into full-stack speech products and agents.

Eridu emerged from stealth in 2026 with a large financing round and a pitch focused on re-architecting AI networking.

Etched is building out Sohu as a transformer-specific inference chip aimed at the most demanding model-serving workloads.

Figure's current public AI story centers on Helix 02 for full-body autonomy, with Go-Big extending internet-scale humanoid pretraining.

Fireworks AI is positioning itself around fast production inference, customization, and model deployment through a developer-first API platform.

Genspark is pushing its AI Workspace as a broader agentic work surface rather than only a search-like interface.

GitHub Copilot keeps expanding from inline assistance toward fuller agent workflows tied to pull requests, planning, and the broader GitHub platform.

Glean is broadening from search into a fuller Work AI platform with assistants, agents, and enterprise graph context.

Groq is continuing to pitch fast, low-cost inference through GroqCloud and its LPU-based hardware stack for production deployment.

H Company is positioning itself around cost-efficient autonomous agentic AI deployed into core enterprise processes.

Harvey is broadening from legal research into a fuller platform layer for document, knowledge, and workflow execution across legal teams.

HeyGen is positioning itself as an AI video generator for text, image, and audio-driven production workflows.

Hippocratic AI is pushing safety-focused healthcare agents for patient engagement and operational workflows.

IBM's current public model story has moved to Granite 4.0, while Granite 3.3 remains active for speech and refined reasoning workflows.

Inception's current model page is led by Mercury 2, with Mercury Edit for coding workflows and Mercury 1 remaining supported for existing customers.

Intercom is expanding Fin as an AI customer service agent inside its broader support platform.

Isomorphic Labs continues building AI drug design programs and pharmaceutical partnerships around its AlphaFold-derived research stack.

Kore.ai is packaging more of its platform around enterprise-ready agentic AI applications instead of only traditional conversational AI.

Krea is broadening from image generation into a fuller creative suite spanning images, video, and 3D.

Lambda is now framing itself around AI factories and large-scale infrastructure for both frontier training and production inference workloads.

LangChain's current platform story centers on LangGraph for agent orchestration and LangSmith for evaluation, observability, and production deployment.

Lepton is now framed around connecting developers to global GPU compute through the NVIDIA DGX Cloud Lepton platform.

Luma is now presenting Uni-1, Ray3.14, and Ray3 as part of a broader push into creative agents that generate, transform, and coordinate media across modalities.

Magic is still centered on frontier code-model research, with ultra-long context work, Google Cloud partnership, and AGI-readiness framing.

Midjourney is now explicitly framing itself as a community-funded research lab building both image and video models, with additional software and hardware projects still to come.

Mistral's current stack spans flagship deployment with Mistral Large 3, reasoning with Magistral, and audio with Voxtral.

Modal is positioning itself as high-performance AI infrastructure for teams bringing their own code and workloads.

Modular is positioning MAX as a unified compute layer for production inference across cloud and edge hardware.

Nebius is positioning itself as an AI-native cloud for building and running models and applications.

Nous Research continues expanding Hermes and its broader open-intelligence research stack, including decentralized training efforts.

Otter.ai is continuing to package transcription, summaries, and action-oriented meeting workflows into an AI assistant product for business teams.

Pathway is now publicly pitching Baby Dragon Hatchling as a post-transformer frontier model for continual learning, with the product still in a waitlist-stage research posture.

Physical Intelligence's latest public model progression now runs from pi0 through pi0.5 to pi*0.6, with FAST as the underlying action-tokenization system.

Pika's current web product is centered on Pika 2.5 for video creation and Pikaformance for audio-synced expressive generation.

Poolside is framing software agents as its enterprise beachhead while building foundation models and orchestration systems around them.

Replicate remains focused on making open-source AI models easy to run through a hosted API.

Replit is leaning into agentic software creation inside a browser-native environment that can move from prompt to app to hosted deployment.

Runpod is continuing to position itself as an AI and cloud infrastructure provider for GPU-heavy workloads.

Runway's newest public lineup now includes Gen-4.5 and its first general world model family, GWM-1, alongside API image generation.

SSI continues to position itself as a straight-shot frontier lab with no product line beyond safe superintelligence.

SambaNova is positioning itself as a full AI platform for fast inference, fine-tuning, and enterprise-scale deployment.

Sarvam AI is framing itself as India's full-stack sovereign AI platform.

Scale AI is now pairing its GenAI Data Engine and evaluation work with broader enterprise and public-sector application infrastructure.

Sierra is scaling branded customer-facing AI agents for major consumer and enterprise companies.

Trying to convert legacy mindshare in open media into a more durable product footprint.

Synthesia is extending from avatar video generation into a fuller enterprise video communications platform with assistant, translation, and publishing workflows.

Together AI keeps expanding its AI Native Cloud with dedicated container inference, broader model-shaping workflows, and enterprise-grade fine-tuning.

Udio keeps iterating on controllable music generation while navigating licensing scrutiny from major rights holders.

Unitree keeps widening its humanoid and legged robot lineup, with H1, G1, and newer consumer-facing launches pushing it further into mainstream deployment.

v0 continues moving beyond one-off UI generation toward a fuller build loop for web products, especially for fast prototyping and iteration.

Vercel's current AI stack now spans the AI SDK, AI Gateway, v0, and agent-oriented cloud primitives aimed at taking apps from prototype to production.

Waabi keeps advancing its simulation-led autonomous trucking platform with Waabi Driver and Waabi World.

Weights & Biases is still centered on the AI developer tooling layer, now with added context from its integration into CoreWeave after the 2025 acquisition.

Windsurf keeps focusing on the handoff between chat, search, and direct code changes, aiming to make agentic coding feel more continuous than episodic.

Wiz continues expanding from cloud posture into broader code, runtime, and AI-related cloud security.

Differentiating by controlling the model layer instead of only wrapping other labs.

You.com is leaning into enterprise AI search infrastructure, APIs, and agent-era search services rather than only consumer search.

01.AI remains centered on the Yi model family and its position inside China's open-model ecosystem.

Abnormal Security continues expanding AI-native detection and response across email and collaboration security workflows.

Adept is positioning its agent stack around resilient workflow automation for enterprise software environments.

AMI Labs has now launched publicly and raised funding around a world-model-focused research agenda spanning planning, memory, and controllable AI.

Apollo.io continues adding AI-assisted prospecting and outbound workflow capabilities around its core sales data platform.

Arcee AI is pushing its open-weight model lineup and enterprise positioning around efficient, deployable models.

Autoscience Institute has started publicly framing itself as an automated AI research lab after raising seed funding in 2026.

Positioning math-native reasoning as a wedge into high-trust scientific and technical workflows, with AXLE emerging as a notable tool in the stack.

Beautiful.ai continues positioning AI-assisted presentation generation as a faster alternative to manual slide design.

Remaining a major upstream source of agent, robotics, and multimodal ideas.

Buildots continues expanding computer-vision-driven progress tracking and performance tooling for large construction projects.

Celestial AI is advancing its photonic fabric platform as a performance and efficiency layer for next-generation AI systems.

Descript is continuing to position itself as an AI-native editor for video and audio workflows.

Edra is positioning itself around converting expert knowledge into scalable playbooks and operational systems.

Eightfold AI continues positioning its talent intelligence platform across hiring, mobility, and workforce planning workflows.

ELSA continues expanding AI-driven pronunciation coaching and spoken-language learning experiences.

Factory is pushing toward agentic software delivery for teams, where multiple coding agents can participate inside a shared workflow rather than a single-chat interface.

Fireflies.ai continues expanding from transcription into AI meeting workflow automation and collaboration features.

Gamma continues expanding AI-assisted document and presentation generation for individual and team workflows.

Trying to translate long-built biological infrastructure into defensible AI leverage.

Graphcore continues to position its IPU stack for machine learning acceleration in cloud environments following its SoftBank-backed reset.

Insilico Medicine continues commercializing AI drug-discovery programs and partnerships around its Pharma.AI stack.

Jasper continues evolving from a writing assistant into a broader enterprise marketing and content workflow platform.

Kana emerged from stealth with a marketing-agent pitch built around loosely coupled AI agents for campaign and audience work.

Trying to establish a European nonprofit alternative to closed frontier labs.

LEAP 71 is pushing computational engineering as a software-first way to generate advanced real-world designs across industries.

Lemonade continues using AI-native operations as a core part of its insurance product and claims experience.

Letta continues productizing stateful agent infrastructure and persistent memory for developers building long-lived AI systems.

Levangie Labs is framing cognitive architecture and episodic memory as the core infrastructure for expert-level autonomous agents deployed with partners.

Lightricks is pushing LTX Studio as a filmmaking platform while also highlighting LTXV as an in-house open-source AI video model.

Liquid has refreshed its lineup around the LFM2 family, including LFM2-24B-A2B and multimodal LFM2.5-VL for cloud-to-edge deployment.

Logical Intelligence is piloting Kona 1.0 and positioning Aleph as the path toward verified reasoning and code in critical systems.

Manus continues positioning itself around autonomous task execution in cloud and browser environments.

Continuing to influence the talent and ideas pipeline for commercial labs.

Nomic is now focusing on domain-specific AI for architecture, engineering, and construction workflows.

OpenClaw is positioning itself as an open agent operating system for personal and enterprise AI workflows.

Paradox continues expanding Olivia as a recruiting automation layer for hiring and candidate engagement.

Recursion's latest public model milestone is Boltz-2, which now sits alongside its broader Recursion OS 2.0 and Virtual Cell platform work.

Reflection AI is positioning itself around frontier open intelligence and agentic systems.

Reka is positioning itself around multimodal, modular intelligence built for real-world tasks and deployment.

Sakana remains research-heavy, but its current public work now spans Continuous Thought Machines, The AI Scientist, and lightweight open releases like TinySwallow.

SiMa.ai is scaling its physical-AI platform across robotics, automotive, industrial automation, and defense use cases.

Building around the thesis that a shared robot brain can generalize across platforms.

Snowflake's model layer now spans Arctic for language, Arctic embeddings for search, and Arctic-TILT for document understanding inside Cortex AI.

Speak continues pushing AI-native conversation practice and language tutoring for global learners.

Influencing the field through benchmarks, analysis, and academic coordination rather than product shipping.

StepFun continues building multimodal and agentic systems in the increasingly crowded China frontier-model market.

Suno's current stack has moved on to v5, with v4.5+ still in the product mix and ReMi continuing as its songwriting assistant.

Supermemory is productizing persistent memory and context infrastructure so AI agents and apps can retrieve and reuse relevant knowledge more reliably.

Taalas is pushing a model-as-chip architecture as a radically more efficient path for inference than standard GPU stacks.

Tenstorrent continues to position itself as a next-generation computing company built for AI.

Tractable continues applying computer vision and workflow automation to insurance claims and related operational assessments.

Typeface is expanding its enterprise content workflow platform around brand-safe generative marketing.

Unreasonable Labs emerged from stealth in 2026 with a platform centered on AI reasoning for scientific discovery.

Vast.ai continues to position itself around low-cost cloud GPU access with flexible global capacity.

World Labs has now productized its work around Marble 0.1-plus, Marble 0.1-mini, and the broader World API.

Yutori is expanding Scouts as an always-on monitoring agent for consumer and research workflows.