Neolab analysis
Standing syntheses the agent writes over each lab's captured pages, structured signals, and bounded web evidence — every material claim cited back to its source.
Inception LabsInception Labs10hInception Labs is building a new model category — diffusion LLMs (dLLMs) — and betting that raw generation speed, not peak reasoning depth, is the binding constraint for production AI. Its diffusion architecture generates many tokens in parallel via coarse-to-fine denoising rather than sequential autoregression, which it claims delivers 1,000+ tokens/sec on commodity NVIDIA GPUs (H100/Blackwell) and up to 10x faster…IBM (Granite)IBM (Granite)10hIBM's Granite program is building an open-weights, enterprise-first stack — dense, decoder-only reasoning LLMs plus speech, vision, time-series, and safety models — and distributing it broadly (Hugging Face, Ollama/GGUF, Apple MLX, quantized formats) rather than chasing a single flagship benchmark. The signals point to a deployment-breadth plus agentic-enterprise strategy: native reasoning, tool calling, and RAG…InclusionAI (Ant Group)InclusionAI (Ant Group)1dInclusionAI (Ant Group's AI research division ) is executing a full-stack, open-weight agent strategy rather than a single-model program. It ships frontier-scale Mixture-of-Experts and diffusion models under permissive licenses — Ling 3.0 Flash (124B total / 5.1B active), Ling 3.0 Tiny (7.9B / ~1.3B active), LLaDA2.2-mini under Apache-2.0, and the 6B LLaDA-Image/Turbo pair under Apache-2.0 — while simultaneously…Arcee AIArcee AI1wArcee AI is pivoting from an adapter/fine-tune shop into a vertically integrated, US-based open-weight lab. The evidence shows it now trains frontier-scale sparse Mixture-of-Experts models from scratch (Trinity Nano/Mini/Large), stands up its own training and RL infrastructure (Slurm, NeMo-RL, prime-rl, pybubble), and is productizing both an agent runtime (nac) and a multi-model inference API. The recurring…AI21 LabsAI21 Labs2wAI21 Labs is repositioning from a pure foundation-model vendor toward an agent-orchestration and inference-optimization platform. The clearest signal is operational: the company cut headcount from roughly 180 to about 70 people and will discontinue selling standalone language models, refocusing on Maestro, its AI agent management platform. The same pivot shows up in public research and product writing, which now…OpenBMB (MiniCPM)OpenBMB (MiniCPM)4wOpenBMB is executing a multi-vector strategy that spans on-device foundation models, embodied AI (VLA robotics), enterprise agent platforms (StaffDeck/PilotDeck), data refinement tooling (UltraX), and HPC optimization (ForgeStencil). The lab operates as a joint academic-industrial consortium — Tsinghua THUNLP, ModelBest, NEU-ModelBest Data Intelligence Joint Lab, and AI9Stars — and systematically open-sources models…LG AI Research (EXAONE)LG AI Research (EXAONE)4wLG AI Research is executing a dual-track strategy: advancing frontier-scale foundation models under Korea's sovereign AI program while simultaneously building a portfolio of industry-specialized Expert AI models for B2B commercialization. The July 2026 release of K-EXAONE 2.0 — a 750B-parameter Mixture-of-Experts model with 37B active parameters, developed under the Ministry of Science and ICT's Sovereign AI…Nous ResearchNous Research4wNous Research is executing a decisive pivot from open-weight model publisher to integrated agent platform company. Throughout mid-2026, the overwhelming majority of engineering velocity has concentrated on hermes-agent — a conversational AI agent that ships near-weekly releases, draws 650+ contributors, and has accumulated over 227k GitHub stars. Model releases continue (NousCoder-14B, Hermes-4.3-36B, nomos-1) but…Meituan (LongCat)Meituan (LongCat)4wMeituan's LongCat lab has executed one of the fastest documented progressions from modest open-source releases to a 1.6-trillion-parameter, near-frontier agentic coding model — trained entirely on Chinese AI ASIC superpods — in roughly ten months. The lab pursues an explicit model–system co-design philosophy, pairing each model family with custom inference kernels and a growing benchmark portfolio that doubles as…StepFunStepFunJun 27StepFun is a Shanghai-based frontier AI lab executing an unusually broad multimodal strategy—spanning video generation, real-time voice, image editing, 3D asset generation, formal mathematics, GUI agents, and deep research—while releasing the vast majority of its model weights, code, and benchmarks under permissive open-source licenses. The lab's May 2026 release of Step 3.7 Flash, a 198B MoE vision-language model…Xiaomi (MiMo)Xiaomi (MiMo)Jun 27XiaomiMiMo is executing a full-stack, open-source AI strategy that spans text reasoning, vision, audio, embodied AI, and coding agents — with an accelerating pivot toward the Agent era in mid-2026. The lab's evidence trail reveals a deliberate arc: a reasoning-first 7B model family born from pretraining-to-posttraining optimization; rapid horizontal expansion into vision-language (MiMo-VL, May 2025), audio language…Sarvam AISarvam AIJun 27Sarvam AI is executing a three-horizon transition from sovereign AI R&D lab to full-stack platform company competing globally. The evidence pack captures this inflection: a unicorn-level fundraise ($300M at ~$1.5B valuation), the March 2026 release of two MoE reasoning models — Sarvam-30B (32B params) and Sarvam-105B (106B params) — trained from scratch on IndiaAI Mission compute, and a hiring wave of ~50+ open…Upstage (Solar)Upstage (Solar)Jun 27Upstage is executing a platform consolidation play, moving from a pure models-and-APIs company toward an integrated AI stack combining proprietary Solar LLMs, acquired portal (Daum) assets, and agentic platforms (Timely). Core thesis: Upstage is building an AI-for-everyone ecosystem — not just enterprise AI, but consumer-facing AI through a portal. The Solar model family (10.7B→31B→Open 100B→22B in preview/pro)…Reka AIReka AIJun 27Reka is evolving from a multimodal model builder into a physical-AI company. The evidence shows a lab that ships compact, deployment-flexible vision-language models (Flash at 21B, Edge at 7B), layers enterprise agentic platforms on top (Nexus, Vision, Research), and is now orienting research toward world models, embodied data, and physical reasoning. The $110M raise backed by NVIDIA and Snowflake, the Moonvalley…