AI Model Launch Tracker

What we captured

HUMAIN launched the 428-billion-parameter humain-m3 Arabic-focused mixture-of-experts model as a research preview through HUMAIN Node, with a planned open-weight release under the MiniMax Community License after safety and alignment work.

• • HUMAIN announced humain-m3 at LEAP Riyadh on September 3, 2026.
• • The model has 428 billion parameters and is based on the MiniMax-M3 lineage.
• • HUMAIN reported an 89.37% average across seven public Arabic benchmarks.
• • The current release is a research preview; open-weight release is planned after safety and alignment work.

Source evidence

Confidence 99% / Completeness 91%

What we captured

Google Research and Google DeepMind launched WeatherNext 3, an hourly global weather model using live satellite and weather-station observations, with availability across Google products and cloud geospatial services.

• • Uses live global geostationary satellite data and weather-station observations.
• • Generates hourly forecasts with surface-variable resolution as fine as 5 kilometers.
• • Google reports up to 50% more accurate precipitation forecasts one or more days ahead.
• • Includes 100-meter wind, cloud-cover, and solar-radiation forecasts for renewable-energy applications.

Source evidence

What we captured

Meta launched the proprietary Muse Spark 1.3 model through Muse Code and the Meta Model API. The shipping xhigh configuration is broadly rolling out, while the higher-performing max configuration remains in limited partner preview pending safety testing.

• • Meta says Muse Spark 1.3 delivers frontier-level coding and agentic performance.
• • The broadly available xhigh configuration scored 61 on Artificial Analysis's Intelligence Index; max scored 62.
• • Standard pricing remains $1.25 per million input tokens and $4.25 per million output tokens.
• • The max configuration is not yet broadly available and is undergoing additional safety testing.

Source evidence

What we captured

Google launched generally available Gemini 3.8 Flash with a one-million-token context window and introductory API pricing, alongside the restricted Gemini 3.8 Flash Cyber variant for trusted defenders.

• • Gemini 3.8 Flash was introduced on September 2, 2026.
• • Introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026.
• • Standard pricing begins January 1, 2027, at $1.50 per million input tokens and $7.50 per million output tokens.
• • Gemini 3.8 Flash Cyber is restricted to trusted defenders through the Fairwind Program.

Source evidence

What we captured

Google released Gemini 3.8 Flash for agentic and software-development workloads and introduced a restricted cybersecurity variant for trusted defenders focused on vulnerability discovery and patching.

• • Google announced two Gemini 3.8 Flash variants on September 2, 2026.
• • The standard variant targets agentic tasks, software development, and multi-step reasoning.
• • The Cyber variant targets vulnerability discovery and automated patching and is restricted to trusted defenders.
• • The standard variant has a 1 million-token input window and costs $0.75 per million input tokens and $3.75 per million output tokens.

Source evidence

What we captured

Muse Spark 1.3 launches in Muse Code and Meta Model API with improvements for agentic workflows, coding, multitasking, and long-form instruction following.

• • The model is available in Muse Code and Meta Model API at launch.
• • Maximum reasoning mode remains pending additional safety testing.
• • The release emphasizes longer-horizon agentic work, multitasking, instruction following, and capability awareness.

Source evidence

Confidence 100% / Completeness 87%

What we captured

Google launched two Gemini 3.8 Flash models, including a general-purpose model available through Google AI Studio and Android Studio and a cybersecurity model restricted to trusted government and critical-infrastructure entities through the Fairwind Program.

• • Google announced both model launches on September 2, 2026.
• • Gemini 3.8 Flash supports multimodal inputs and has a 1,048,576-token input limit.
• • Gemini 3.8 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens.
• • Gemini 3.8 Flash Cyber targets vulnerability detection and automated patching and achieved a stated 47.2% pass rate on CWE-Bench.

Source evidence

What we captured

Anthropic announced Claude Mythos 5.1, available to vetted cybersecurity and life-sciences organizations through trusted access programs.

• • Announced September 1, 2026.
• • Available through trusted access programs for vetted cybersecurity and life-sciences organizations.
• • Pricing starts at $10 per million input tokens and $50 per million output tokens.

Source evidence

Confidence 99% / Completeness 92%

What we captured

Runway announced Solaris, a research-stage Interface World Model that generates and dynamically updates software interfaces in response to user interaction.

• • Solaris generates interfaces frame by frame rather than executing conventional interface code.
• • The system responds to clicks, drags, and voice commands and renders at 720p.
• • Runway says text rendering, reliability, long sessions, and screen-reader compatibility remain unresolved.

Source evidence

What we captured

Anthropic released Claude Fable 5.1 for general availability and Claude Mythos 5.1 for restricted access, with a 1M-token context window, 128K maximum output, and a 75% reduction in cache-read pricing.

• • Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, versus 24.7% for Fable 5.
• • Cache reads were reduced from $1.00 to $0.25 per million tokens.
• • The two releases are described as one underlying model with different safeguard layers.

Source evidence