This observation covers 7:12 am on 12 October to 7:12 am on 13 October 2026 in Sydney, Australia. It includes two technical changes that were formally announced and can be verified from primary material.
Meta opens access to Muse Wonder and Muse Wonder Full
Meta launched Muse Wonder and Muse Wonder Full on 12 October. Muse Wonder 1.0 is now available through the Meta Model API and Muse Code. Meta's developer page lists a one-million-token context window and API prices of US$4 per million input tokens, US$0.20 for cached input and US$20 for output. Muse Wonder Full 1.0 is not available through the general API; access is currently limited to the Power Usage and Maximum Usage Muse Code subscription plans. Both are proprietary hosted models, and Meta has not disclosed their weights, parameter counts or training-data composition.
The change from the earlier Muse Spark release is that the larger models have entered usable services, accompanied by descriptions of Active Memory and Active Focus for longer-running work. Meta says the models can operate across applications, sustain multi-step tasks, delegate to sub-agents and recover from errors. Muse Wonder is also intended to power Meta's Muse personal agent, but that integration is not yet complete and no date for broad availability has been published. Meta's benchmark table covers legal, finance, health, computer-use, memory and long-horizon software-engineering tasks. Most results are developer-reported; Meta says DeepSWE and SWE-Marathon include independent third-party runs, but its full safety-evaluation report remains forthcoming and there is not yet broad independent replication of all capability claims.
Direct source: Meta AI Research, “Muse Wonder”, 12 October 2026, https://research.meta.ai/blog/muse-wonder
Meta developer platform, “Muse Wonder”, 12 October 2026, https://dev.meta.ai/models/muse-wonder
Reuters, “Meta releases Muse Wonder AI models to challenge Anthropic and OpenAI”, 12 October 2026, https://www.reuters.com/business/meta-releases-muse-wonder-ai-models-challenge-anthropic-openai-2026-10-12/
Microsoft discloses the Maia 300 inference-system architecture
Microsoft disclosed the third-generation Azure Maia inference system, Maia 300, on 12 October. Production deployment is scheduled to begin in 2027, so its current status is a formal announcement and technical disclosure rather than an operating deployment. The accelerator uses a multi-chiplet architecture, with compute chiplets manufactured on TSMC's N2P process. Each accelerator has eight HBM4 stacks providing 288 GB of capacity and 18 TB/s of peak bandwidth within a 950-watt thermal design power. A complete rack contains 96 accelerators in a 167-kilowatt power envelope, with 27.6 TB of aggregate HBM capacity, and the two-tier interconnect can extend to a 12,288-accelerator superpod.
Microsoft reports that Maia 300 provides up to seven times the rack-scale token throughput and more than 2.5 times the performance per watt of Maia 200, and expects it to deliver 40 per cent more tokens per megawatt than leading GPU-based systems. These comparisons are Microsoft-reported, the comparison systems were not named, and production hardware has not yet been independently benchmarked. Microsoft already uses Maia 200 for production workloads and says it is deploying that system across the United States, Europe and Australia. Maia 300 is planned for multi-gigawatt expansion beginning in 2027 for long-context reasoning, latency-sensitive conversation and sustained multi-step agent workloads. The architecture, rack specifications and deployment plan are confirmed; actual production volume, delivery timing and third-party performance remain unverified.
Direct source: Microsoft Official Blog, “Microsoft to expand Azure Maia system to multi-gigawatt deployments”, 12 October 2026, https://blogs.microsoft.com/blog/2026/10/12/microsoft-to-expand-azure-maia-system-to-multi-gigawatt-deployments/
Microsoft Azure Blog, “Deep dive into Azure Maia 300 architecture”, 12 October 2026, https://azure.microsoft.com/en-us/blog/deep-dive-into-azure-maia-300-architecture/
Reuters, “Microsoft plans huge deployment of its own AI chips, with an eye on saving costs”, 12 October 2026, https://www.reuters.com/business/microsoft-plans-huge-deployment-its-own-ai-chips-with-an-eye-saving-costs-2026-10-12/
Sources
Meta AI Research, “Muse Wonder”, 12 October 2026, https://research.meta.ai/blog/muse-wonder
Meta developer platform, “Muse Wonder”, 12 October 2026, https://dev.meta.ai/models/muse-wonder
Reuters, “Meta releases Muse Wonder AI models to challenge Anthropic and OpenAI”, 12 October 2026, https://www.reuters.com/business/meta-releases-muse-wonder-ai-models-challenge-anthropic-openai-2026-10-12/
Microsoft Official Blog, “Microsoft to expand Azure Maia system to multi-gigawatt deployments”, 12 October 2026, https://blogs.microsoft.com/blog/2026/10/12/microsoft-to-expand-azure-maia-system-to-multi-gigawatt-deployments/
Microsoft Azure Blog, “Deep dive into Azure Maia 300 architecture”, 12 October 2026, https://azure.microsoft.com/en-us/blog/deep-dive-into-azure-maia-300-architecture/
Reuters, “Microsoft plans huge deployment of its own AI chips, with an eye on saving costs”, 12 October 2026, https://www.reuters.com/business/microsoft-plans-huge-deployment-its-own-ai-chips-with-an-eye-saving-costs-2026-10-12/
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.