Provenance Record

Strong, well-structured News piece (706 words, within the 400-1200 News range). Clear Overview / What We Know / What We Don't Know / Analysis structure. Technical specifics are precise and attributed inline. The 'What We Don't Know' section responsibly flags that benchmark figures are Microsoft's own and that independent third-party evaluations had not been released, and that the OpenAI agreement terms were not fully disclosed.

All 5 source snapshots read from disk (gunzip), all fetched at status 200. source-0 (microsoft.ai hillclimbing): confirmed the seven model names, the 'long term self-sufficiency for Microsoft and our partners' quote, 'We train our reasoning models from scratch. We don't distill from other labs and we don't rely on unlicensed or opaque data', MAI-Code-1-Flash '5 billion active parameters' + 'tailor-made for and deeply integrated into GitHub Copilot, VS Code and the Microsoft stack', 'Maia 200 silicon' + '1.4x efficiency boost', 'hill-climbing machine' quote, 'Humanist Superintelligence', and OpenRouter/Fireworks/Baseten — all verbatim. source-1 (microsoft.ai MAI-Thinking-1): confirmed '35B-active, ~1T-total parameters, sparse Mixture of Experts model', '256k token window', 'reaches 97.0% on AIME 2025, and 94.5% on AIME 2026', 'toe-to-toe with Claude Opus 4.6 on SWE-Bench Pro', 'users preferred MAI-Thinking-1 over Claude Sonnet 4.6', 'trained without distillation from third party models', 'from the ground up on clean, traceable and enterprise-grade data', and 'available in private preview on Microsoft Foundry today' — all verbatim. source-2 (IndexBox): confirmed 'surpassed OpenAI's GPT 5-5 while achieving a tenfold reduction in costs' (note: IndexBox itself prints the typo 'GPT 5-5'; the article reproduces it faithfully inside quote marks), the revenue-sharing ceiling / 'terminating its exclusive right to market OpenAI's models' quote, and the Fireworks/Baseten/OpenRouter availability. source-3 (Technobezz): confirmed '35 billion active parameters', '256,000-token context window', 'trained from scratch on Azure infrastructure using commercially licensed data, with no distillation from third-party systems', 'outperformed OpenAI's GPT-5.5 on quality with what it projects as ten times better cost efficiency', the '$13 billion ... across multiple tranches beginning in 2023, securing exclusive cloud rights on Azure' line, and that Nadella took the Build stage on June 2. source-4 (GIGAZINE): confirmed 'On June 2, 2026, Microsoft announced seven of its proprietary AI models' verbatim and that GIGAZINE independently lists the same seven models (with minor punctuation differences such as 'MAI Transcribe-1.5' and 'MAI-Image-2.5 Flash').

Every direct quote is verbatim from its cited source. Benchmark numbers (97.0% AIME 2025, 94.5% AIME 2026) and parameter counts (35B-active / ~1T-total sparse MoE, 256k context, 5B for MAI-Code-1-Flash) trace verbatim to the two official Microsoft pages, satisfying Rule 9 (benchmark claims cited to the primary source). Speaker attribution is correct: the Suleyman quotes are tied to the Microsoft AI blog post; Nadella is referenced as taking the Build stage via Technobezz. Business specifics ($13B investment, GPT-5.5 'tenfold cost', revenue-sharing ceiling) are kept OUT of the lead/headline/summary and live in the Analysis section, each attributed to Technobezz and/or IndexBox. The lead does not rest on any single non-allowlisted source. CNBC and Windows Central are correctly NOT cited.

High-quality, accurate, well-sourced News submission. All quotes verbatim, all specifics traced to cited sources, correct attribution, original topic. Approved with a corrections record documenting the allowlist warning and the faithfully-reproduced 'GPT 5-5' source typo.