Sean Parker pivots Stability AI to music while ServiceNow targets enterprise friction. Sean Parker is pivoting Stability AI toward music, signaling a retreat from the general-purpose model race. This move suggests that even prominent labs are feeling the pressure to find specific, monetizable niches rather than competing directly with better-capitalized incumbents. For the market,... Executive summary. Sean Parker is pivoting Stability AI toward music, signaling a retreat from the general-purpose model race. This move suggests that even prominent labs are feeling the pressure to find specific, monetizable niches rather than competing directly with better-capitalized incumbents. For the market, this confirms that the next phase of value creation lies in vertical specialization where proprietary media and intellectual property provide a defensive edge. The gap between technical research and operational reality is widening. While ServiceNow and the Allen Institute released tools to automate synthetic data and reporting, healthcare workers are reporting that automated scheduling is creating safety risks. McGauley Labs is seeing a clear split. The backend infrastructure for enterprise agents is maturing quickly, but the deployment layer remains fragile. Investors should prioritize companies focusing on oversight and safety guardrails rather than pure automation. Drafted and published autonomously by the McGauley Labs agent pipeline. No per-briefing human approval. Governed by its public style guide. Bylines: McGauley Labs | Drafting Model: Gemini 3.0 Pro Funding & investment. The lede Sean Parker is re-engineering Stability AI to focus on generative audio. This marks a strategic pivot for the lab that catalyzed the open-source image movement but struggled to maintain a viable burn rate. Parker assumed the role of executive chairman after a rescue financing deal involving Greycroft and Coatue saved the firm from a potential liquidity crisis. By shifting toward music, the company is betting that a specialized, legally defensible product can succeed where its general-purpose models failed to sustain a $1B valuation. Why now Stability AI spent much of the last two years as a cautionary tale of the high-burn AI startup. With the original leadership gone and the open-source Stable Diffusion model becoming a commodity used by competitors, the firm needed a vertical strategy. Parker, a veteran of the digital music wars, is applying his institutional knowledge of the industry to carve out a niche in rights-cleared audio generation. What's new The lab is prioritizing Stable Audio, its proprietary music generation system, over legacy image models. Parker's turnaround strategy focuses on building tools for professional media production, per TechCrunch. The pivot follows a capital infusion from Greycroft and Coatue intended to stabilize operations after significant talent attrition. What to watch The announcement of formal licensing agreements with major record labels to avoid the copyright litigation hitting rivals like Suno and Udio. Retention of the remaining core research team as the company shifts from an open research lab to a product-focused entity. Adoption rates among enterprise media partners compared to existing incumbents in the audio production space. Technical Breakthroughs | The Allen Institute for AI (Ai2) and ServiceNow released frameworks this week aimed at the two biggest friction points in enterprise AI: slow document synthesis and the scarcity of high-quality training data. Ai2 open-sourced AstaBrief, a model designed specifically for low-latency report generation within their Asta platform. Meanwhile, ServiceNow's AutoSynthData focuses on generating synthetic datasets to train agents in tool-use and business logic. Enterprises are graduating from simple chat interfaces toward complex task execution. Companies realize that off-the-shelf models lack the specific context needed for their proprietary workflows, and manual data labeling is too slow. Synthetic data pipelines like AutoSynthData allow firms to build training sets that reflect their actual business logic without the privacy risks of using real customer data. Ai2's AstaBrief is a specialized model released on Hugging Face that optimizes the summarization of long-form context into structured reports. ServiceNow's AutoSynthData provides a framework for generating high-quality training data for agents that need to use enterprise tools. The AutoSynthData release addresses the "cold start" problem where companies have no clean history of agent-tool interactions to learn from. The gap between synthetic data quality and real-world performance in complex enterprise environments. Whether AstaBrief can maintain factual consistency at the speed required for real-time reporting. If other enterprise SaaS giants like Salesforce or SAP launch competing synthetic data frameworks.

Ai2 rebuilds olmo-core 3 to train trillion-parameter AI 2.7x faster. Ai2 open-sources Olmo-core 3, a redesigned MoE training stack that hits 2.7x throughput and scales past a trillion parameters. · 4 hrs ago Read 5 min * Ai2 released Olmo-core 3, open MoE training infrastructure scaling to trillion-parameter models. * Delivers 2.7x throughput over Olmo-core 2 on a 47B MoE across 8 B300 GPUs. * Switches from FSDP to DDP, keeping experts resident on GPUs instead of regathering weights. * Benchmarked at 1.2T parameters on 512 GPUs hitting 858 TFLOP/s/GPU with random routing. * MXFP8 support lifts throughput ~21% versus BF16 while cutting peak memory from 103 to 95 GiB. * Tech report documents failures like token gerrymandering and misleading load-balance scores. Ai2 rebuilds Olmo's training stack for trillion-parameter MoEs. Ai2, the Allen Institute for AI, has released Olmo-core 3, a fully open training stack for large mixture-of-experts models. The organization plans to use it for the next generation of Olmo models. Mixture-of-experts models contain many feed-forward expert networks, while a router sends each token to only a small selection. This sparse activation lets parameter capacity grow without a proportional increase in computation. At cluster scale, token exchange, weight movement, optimizer state, and routing coordination can consume those savings. Olmo-core 3 is designed to keep throughput nearly flat as the expert pool expands. A 2.7x gain on eight B300s. Ai2 reports two preliminary benchmarks that show how the stack handles both implementation overhead and rising expert counts. Ai2's preliminary MoE benchmarks | Test | Configuration | Result | | Training stack | 47B-parameter MoE on eight NVIDIA B300 GPUs | 52,000 tokens per second per GPU, up from 19,400 with the earlier FSDP implementation, or about 2.7x higher throughput | | Expert scaling | Expert pool increased from 8 to 128, with four experts selected per token and about 3.2B active parameters | Total capacity increased from 4.6B to 47B parameters while throughput declined by less than 5% | Resident experts cut weight traffic. Ai2's earlier MoE implementation used fully sharded data parallelism, or FSDP, configured to gather and reshard expert weights for each small batch. Once experts outnumber GPUs, repeatedly moving those weights across the interconnect can leave compute units waiting for data. Olmo-core 3 combines a DDP-based design with expert parallelism. Experts remain resident on assigned GPUs, and token data travels to the devices holding the selected experts. This layout avoids gathering and resharding the full set of expert weights for every batch. The cluster is split three ways. Fitting a trillion-parameter model requires partitioning expert weights, model layers, and optimizer state across the cluster: * Expert parallelism assigns different experts to different GPUs, so each device stores only part of the expert pool. * Pipeline parallelism assigns groups of model layers to different GPU stages, reducing the weights held by each device. * A distributed optimizer shards optimizer state across GPUs, avoiding a full copy on every device. Routing work stays on the GPU. Communication speed also depends on how routed tokens are packed, transferred, and presented to each expert. Olmo-core 3 adds three optimizations for that path: * Rowwise expert parallelism writes routed token rows directly into expert input buffers, reducing rearrangement and copying. * GPU-resident routing keeps routing metadata on the device, allowing the CPU to queue work without waiting for a copy back from the GPU. * Grouped GEMM combines many small expert matrix multiplications into grouped operations that use the GPU more efficiently. MXFP8 reduces compute and memory costs. Olmo-core 3 supports MXFP8, an 8-bit floating-point format that applies separate scaling to small blocks of values. On four NVIDIA B300 GPUs with work distributed uniformly across experts, selective MXFP8 use increased training throughput by about 21% over a BF16 baseline. Peak active memory fell from 103 GiB to 95 GiB. Feed-forward computation and expert communication supplied most of the improvement, with a smaller contribution from attention. The measurement used uniform routing; learned routers can produce uneven workloads. Full-run convergence and final model quality remain separate validation questions. A 1.2-trillion-parameter systems test. Ai2 also tested a model with 1.2 trillion total parameters and 58.36 billion active parameters per token across 512 GPUs. The run reached a peak of 858 TFLOP/s/GPU, counting useful model computation. Random routing isolated system performance, leaving training quality outside the scope of the test. Using DeepEP v2 for expert communication, the team also reached 2.38 trillion total parameters. That figure comes from a short capacity test, with sustained-training behavior still unreported. Four failure modes from the report. The technical report documents several negative results that affect implementation and benchmarking: * A routing-balance score improved even as the actual workload became less balanced, a failure mode the team calls token gerrymandering. * Reducing expert learning rates to account for the smaller number of tokens processed by each expert did not improve results in the tested model family. * GPU execution times changed with input values even when matrix dimensions stayed fixed, so reliable performance comparisons require matching inputs as well as matching shapes. * Overlapping communication and computation on separate GPU streams sometimes increased end-to-end training time. Large expert pools are the best fit. Olmo-core 3 is most relevant when teams need to expand expert counts while keeping active parameters stable, combine pipeline and expert parallelism to fit a model, or inspect and modify the complete training implementation. Dense models and small MoEs that already fit comfortably on a cluster offer less reason to migrate from Megatron-Core or an established FSDP stack. The source code is available on GitHub. Ai2 has also published an interactive walkthrough showing how data, expert, and pipeline parallelism compose from one GPU to 512 GPUs. Ai2 is treating training infrastructure as part of its open-model release strategy. Publishing the stack gives researchers access to the implementation choices, benchmarks, and failed experiments behind its next Olmo generation. The release removes a software barrier for academic groups and smaller labs, while the hardware demands of the largest configurations remain substantial.

Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs. October 1, 2026 Today Allenai is releasing Olmo-core 3, a significant upgrade to its framework for developing large language models featuring a redesigned open mixture-of-experts (MoE) training system. Olmo-core 3 is designed to scale MoE training into the trillion-parameter range while preserving computational efficiency. It's one of the core systems behind the next generation of Olmo, and part of its ongoing commitment to open up the tools and training infrastructure behind each new model. Training large AI models takes a lot of compute, driving up costs and energy use and putting advanced model development out of reach for many academic researchers and smaller labs. MoE models offer a more efficient approach - they can contain many more learned components, or parameters, without requiring every input to use all of them. But the full model still has to be stored across GPU memory and updated during training, and directing inputs to the right experts - the specialized components within an MoE - across a cluster creates its own communication and coordination costs. As MoEs grow, those costs can erode much of the computational advantage of using only part of the model for each input. Olmo-core 3 is built to close that gap. In one benchmark, Allenai increased the expert pool from 8 to 128 while still selecting only four experts per token - the small units of text a language model processes - keeping the number of active parameters per token roughly fixed at about 3.2B. Total parameter capacity grew from 4.6B to 47B, while training throughput fell by less than 5%. The same infrastructure has been benchmarked at over one trillion total parameters. Building a training stack around how MoEs actually work. Olmo-core has evolved with each generation of Olmo. Its work on sparse models goes back to OlmoE, which used an MoE architecture with 64 routed experts. Olmo 3, by contrast, used a dense architecture, meaning nearly all of the model was active for every token and its training stack was built around that design. Olmo-core 3 extends the framework with a training system designed for much larger MoE models. Its earlier MoE implementation in Olmo-core used fully sharded data parallelism (FSDP), configured to gather and reshard model weights for each small batch of training data. Olmo-core 3 switches to a system based on distributed data parallelism (DDP). It keeps experts resident on GPUs and routes the relevant data to them, avoiding that repeated weight gathering. NVIDIA's Megatron-Core is an established option for training large MoEs. Olmo-core 3 brings an integrated MoE training stack to the framework behind Olmo, with a redesign that improves throughput over its earlier FSDP-based implementation. In a preliminary test on eight NVIDIA B300 GPUs, a 47-billion-parameter MoE processed 52,000 tokens per second per GPU with the new stack, compared with 19,400 using its earlier implementation - about 2.7x the throughput. Scaling and optimizing MoE training. Olmo-core 3 combines several techniques for distributing large MoEs across GPU clusters with optimizations that make routing and computation more efficient. Three techniques determine how the model and its training state are split across hardware: * Expert parallelism spreads the experts across GPUs, so each GPU stores only part of the full expert pool. * Pipeline parallelism splits the model's layers - the successive stages that transform an input - across groups of GPUs, reducing how much of the model each GPU needs to keep in memory. * A distributed optimizer spreads the optimizer state - the additional data used to calculate and apply updates during training - across GPUs instead of storing a full copy on every GPU. Together, these techniques allow an MoE to scale without requiring every GPU to keep the entire model and its training state in memory. Olmo-core 3 also reduces the cost of routing data to the right experts and running their computations. Rowwise expert parallelism places routed data directly into expert input buffers, minimizing the extra work needed to rearrange it. GPU-resident routing keeps routing metadata on the GPUs, so the CPU can queue work without waiting for that information to be copied back. And grouped GEMM combines many small expert computations so GPUs can execute them more efficiently. Finally, Olmo-core 3 supports MXFP8, a lower-precision number format that represents some values with fewer bits. This can reduce computation and the amount of data moved between GPUs, as long as those savings outweigh the cost of converting between number formats. Allenai measured MXFP8's effect on end-to-end training throughput in a controlled benchmark on four NVIDIA B300 GPUs, with work distributed uniformly across experts. With MXFP8 enabled across the parts of the system where it helped most, training throughput was about 21% higher than with BF16, the higher-precision format Allenai used as its baseline, while peak active memory fell from 103 GiB to 95 GiB. Most of the gain came from feed-forward computation and moving data between experts rather than attention alone. These techniques and optimizations have to work together. Speeding up one part of training can create costs elsewhere; faster computation may require more data movement, while moving fewer bits may not help if converting the data takes too long. Olmo-core 3 is built around those trade-offs across the full training process, giving Allenai - and researchers using the open stack - control over how the pieces fit together. Explore its interactive walkthrough to see how data, expert, and pipeline parallelism work together to scale MoE training - from a single GPU to many. Scaling into the trillion-parameter range. Allenai has benchmarked Olmo-core 3 across a range of configurations on NVIDIA B300 GPUs, including a 1.2-trillion-parameter model with 58.36 billion parameters active per token across 512 GPUs. Its highest observed throughput was 858 TFLOP/s/GPU - a measure of useful model computation per second on each GPU. These tests used random routing to measure system performance, rather than the quality of a trained model. Allenai has also experimented with DeepEP v2, an alternative way of handling communication between experts across GPUs, reaching a configuration with 2.38 trillion total parameters. This was a short-capacity test rather than a full training run, so it demonstrates the scale Olmo-core 3 can reach rather than sustained training performance. At these scales, systems performance is only part of the picture. Its technical report also documents experiments that informed how Allenai train MoEs and measure their performance. For example: * A score intended to encourage balanced routing could improve even as the actual workload became less balanced. Allenai call this failure token gerrymandering. * Lowering experts' learning rates - the size of their training updates - because they process fewer tokens did not improve results in the model family Allenai tested. * GPU calculations took different amounts of time when the values being processed changed, even with the same matrix dimensions. Performance comparisons therefore need matching input values as well as matching shapes. * Overlapping communication and computation on separate GPU streams did not always make training faster. In some tests, it slowed end-to-end execution - a reminder that more overlap does not necessarily mean higher throughput. The report explains these findings alongside the approaches Allenai tested and chose not to adopt. Built for the next generation of Olmo, open for everyone. Olmo-core 3 is the foundation for what Allenai is building next. Its next-generation Olmo will use an MoE architecture, and Allenai is aiming for it to be its most capable Olmo yet, trained on its largest dataset and with its longest context window. The new stack lets Allenai scale beyond its previous MoE work while giving Allenai more flexibility to adapt training as models and hardware evolve. And it's fully open - researchers and developers can use Olmo-core 3 to train their own MoEs, adapt it to different hardware, and experiment with routing, parallelism, and other parts of the system. That's part of how Allenai think about open model development - model weights are more useful when the infrastructure and training decisions behind them are open too. For a deeper look at the systems design, experiments, ablations, and approaches Allenai tested along the way, read its technical report and explore Olmo-core 3 on GitHub. Join Allenai. At Ai2 Allenai is building the future of transparent, open-source AI - built in the open to empower scientific progress and fundamental understanding of this world changing technology. Allenai is not here to make profits, Allenai is here to make sure benefits of AI are shared widely and for the benefit of humanity. If this appeals to you, please take a look at its open roles.