AI Intel Report — Independent intelligence on AI & emerging technology.

Frontier Models

Anthropic Details Chinese Labs Distilling Claude Models in September 2026 Threat Report

The September 10, 2026 report covers misuse disrupted from December 2025 to August 2026, including seven China-based labs targeting Claude through distillation and a fourth cybersecurity incident with an early Claude Opus 4.6 version.

5 MIN READ

Frontier Models

OpenAI Releases GPT-Live-1 Full-Duplex Voice Model in API

The new model collapses separate ASR, TTS and LLM components into one architecture that supports simultaneous listening and speaking, with API access, evals and pricing now public alongside expanded ChatGPT Voice availability.

5 MIN READ

Frontier Models

DeepSeek V4.1 Flash Releases with CSA2 KV Sharing and 40 Intelligence Index Score

The 552B-parameter model introduces CED architecture, Engram memory and DSpark decoding to cut KV cache costs fourfold while outperforming its V4 Pro predecessor on benchmarks and triggering its retirement.

4 MIN READ

Inside a vast industrial data center hall with polished concrete floors and exposed metal ceiling beams a team of anonymized technicians wearing plain gray coveralls and safety vests works with their backs turned to the viewer installing rows of new dense server racks that represent the efficient DeepSeek V4.1-Flash MoE release the racks feature matte black chassis with visible high-density compute modules dense bundles of fiber optic cables and liquid cooling pipes snaking between units symbolizing the 552B parameter scale and mixture of experts architecture one rack section shows integrated vision processing hardware with attached camera arrays capturing and processing real world objects like circuit boards and mechanical components on nearby workbenches illustrating native vision capabilities another area displays extended memory modules and storage arrays evoking the 1M context length with long reels of data tapes and stacked solid state drive trays being swapped in by the technicians older bulkier server cabinets labeled only by their physical size differences are being disconnected and rolled away on dollies representing the previous V4-Pro hardware being phased out the floor is covered with organized cable management trays and anti static mats while ventilation grilles and industrial fans maintain airflow around the equipment a distant supervisor stands near a whiteboard covered in non textual diagrams of performance graphs and cost curves the entire scene emphasizes operational efficiency and cost reduction through the compact modern hardware footprint compared to the bulkier legacy units being removed the lighting comes from overhead industrial fixtures casting even illumination across the metallic surfaces and the concrete pillars separating the rack rows technicians handle tools such as torque wrenches and cable testers while one figure examines a handheld diagnostic device showing abstract waveform patterns without any readable characters the background reveals additional rows of identical new racks stretching into the distance with occasional glimpses of external windows revealing a gray industrial exterior the composition focuses on the tangible hardware transition the physical act of deployment the cooling infrastructure the cable density and the human labor involved in bringing online the lower cost higher performance system all elements remain strictly real world photographic without any overlaid graphics symbols or markings — Illustration: AI Intel Report

Frontier Models

DeepSeek V4.1-Flash MoE Model Undercuts V4-Pro on Benchmarks and Cost

The 552B-parameter release ships with native vision, MIT weights, 1M context and Flash-tier pricing while routing legacy Pro traffic to the new endpoint from mid-September.

By Marcus Vance · 4 MIN READ

Frontier Models

DeepSeek V4.1-Flash MoE Model Undercuts V4-Pro on Benchmarks and Cost OpenAI Releases GPT-Live-1 Full-Duplex Voice Model in API DeepSeek V4.1 Flash Releases with CSA2 KV Sharing and 40 Intelligence Index Score OpenAI Launches GPT Image 2.5 Flare and Sunburst Models