Source Verification: All ten snapshots read from disk (gunzip + text extraction); sha256 of each decompressed snapshot re-computed and matches manifest.json; all ten status 200, no archive fallback, suspicious_patterns null for all. All hosts already in config/source_allowlist.txt (blog.google, cloud.google.com, docs.cloud.google.com, deepmind.google, the-decoder.com, thenextweb.com, artificialanalysis.ai); no additions needed. [0] blog.google (Nov 05, 2026; Anish Nangia, Product Manager, Google DeepMind): CONFIRMED 'generally available across the Gemini API and Gemini for business' (quote verbatim), 'dropped the price 30% from Gemini Omni 1.1 making it available at $0.07 per second for 720p video', 'up from 10 seconds with Omni 1.1', 3-15 s in one-second increments, motion clips up to three seconds, 'audio cues', draft at 360p/720p then upscale to 1080p/4K, Provisioned Throughput, SynthID, 'fast sports action, rapid script changes and complex physics' (verbatim), Google Vids 15-second clips at no additional cost, Gemini app rolling out to Google AI Ultra subscribers. The 1.1 vs 1.2 comparison is example captions only; no preference study or benchmark table (confirmed). [1] Google Cloud blog (November 6, 2026): CONFIRMED 'Today, Gemini Omni 1.2 Flash is generally available on Google Cloud', '$0.07 per second of 720p video, 30% less than Gemini Omni 1.1 Flash', 'Note: Audio references are gated behind allowlisting.' (verbatim). Same post also says 'coming soon to Gemini for business' (see finding; clarification filed). [2] Vertex model documentation (page footer 'Last updated 2026-11-07 UTC'): CONFIRMED model ID gemini-omni-1.2-flash; 'Launch stage: GA'; 'Release date: November 5, 2026'; 'Retirement date: November 5, 2027 or later'; Interactions API Supported, GenerateContent API Not supported; Provisioned Throughput Supported; Batch inference and Context caching Not supported; Audio modality 'Not supported'; output duration '3 to 15 seconds (3 to 10 seconds for first and last frames)' and extension 3 to 10 seconds; aspect ratios 16:9, 9:16. [3] Vertex pricing page: raw HTML is server-rendered (no script needed): the Gemini Omni table is in the stored snapshot and a live curl with a browser User-Agent on 2026-11-09 returned the same table (HTTP 200, 2,498,838 bytes, sha256 ec159279a7f8...; the page differs byte-wise from the snapshot, snapshot sha256 a72f5097d4ad... is the committed one, table text identical). Rows: 'Gemini Omni 1.2 Flash' input $1.50, text output $9.00, Video Output** $12.00; 'Gemini Omni Flash, Gemini Omni 1.1 Flash' input $1.50, text output $9.00, Video Output** $17.50; footnote 1931/5792/8688/17376 tokens per second at 360p/720p/1080p/4k, 32 tokens per audio second. CONFIRMED. No region or date column on the page; Standard pay-as-you-go table, 'Show discount options' toggle off by default. The $17.50 is the page's current listed price for the 1.1 row (same unit, same output type, same page as the $12.00), not a historical capture; an Archive.org capture of the pre-launch page was attempted (CDX lists captures to 2026-10-29) but the archive returned HTTP 429, so the prior-price history is not independently confirmed beyond Google's '30% less than Gemini Omni 1.1 Flash' statements; the headline compares like with like and the article does not claim more. [4] DeepMind model card (Published 5 November 2026): CONFIRMED 'is based on Gemini Omni Flash', architecture/limitations/evaluation deferred to the Gemini Omni Flash card, no benchmark scores, 'capable of changing people's speech ... For now, we are restricting this capability' (verbatim). The card's Inputs line lists audio (see finding). [5] The Decoder (Matthias Bastian, Nov 5, 2026, 'Source: Google'): relays Google's announcement; confirms GA wording, 15 s, 30% / $0.07, Ultra-subscriber Gemini app; derivative, adds nothing independent. Anthropic appears only in its sidebar headlines. [6] The Next Web (Ana Maria Constantin, November 5, 2026, 8:03 pm GMT): CONFIRMED 'down from $0.10 a second in June', 'ByteDance's Seedance 2.5, unveiled in June, makes 30-second clips', and 'In Google DeepMind's own comparisons, testers preferred 1.2 over 1.1 for visual consistency 62% of the time'. The 62% figure appears in no other snapshot (grep of all ten: only TNW), as the article says; the article attributes it to TNW's report of DeepMind's comparisons and states the Herald did not find the source. TNW adds context (prior price, competitor, I/O history, a ranking remark the article does not use) but no independent testing: partial independence only. [7][8][9] Artificial Analysis (text-to-video, image-to-video, video-editing): CONFIRMED Gemini Omni Flash 1.1 text-to-video rank 12 (range 9-12), Elo 1128 ±13, 7,932 samples, $9.12/min; image-to-video rank 16 (range 15-17), 1044 ±14, 6,478 samples; video-editing rank 4 (range 2-4), 1146 ±10, 8,271 samples ($9.64/min there). No 1.2 entry on any of the three (searched all pages for 'Omni' and '1.2'); the 'Elo from pairwise human preference votes' and '(1080p where offered)' quotes verbatim. The 'no independent score' statement is scoped to those three boards as read on November 9, which is accurate. Claim-accuracy tests: (1) absence/quantifier grep run on title, summary and body: 'Every capability claim below is Google's unless another source is named' holds (Herald arithmetic and AA/TNW items are labelled); 'None of those three pages lists a 1.2 entry' holds for each; 'None of the Google pages ... explains the difference' stands (the Cloud blog gives the gating fact but does not link it to the doc's 'Not supported'; the article reports it without reconciling, per policy). The 62% absence claim across Google's five pages verified. (2) Headline/summary/lead: qualifiers kept ('Google says', 'as the Herald read them'); GA wording matches Vertex ('Launch stage: GA'), the Gemini API (blog.google) and the Cloud blog; the Gemini app is correctly described as rolling out to Ultra subscribers only; '15-Second Clips' is the maximum output duration for generation and editing (3-15 s; first/last-frame and extension modes cap at 10 s, which the body states); '$12 per Million Tokens' is the Video Output SKU, '$17.50' the Omni 1.1 row. (3) Dates: Nov 5 (blog, doc release date, card), Nov 6 (Cloud blog) as printed; article says so. (4) Derived figures recomputed: 5792 x $12/1e6 = $0.069504/s; 5792 x $17.50/1e6 = $0.10136/s; 1 - 12/17.5 = 31.43% (article 31.4%); rounded $0.10 to $0.07 = 30%; 8688 tokens/s x 60 = 521,280 tokens: $6.255 at $12 (article $6.26) and $9.12 at $17.50 (equals AA's $9.12/min); 15 s at 720p = $1.0426 (article $1.04); all labelled the Herald's own. Vendor figures are labelled Google's. (5) Disagreements: audio input (doc 'Not supported' vs pricing page and card vs blog/Cloud blog gating) reported with attribution; Gemini for business wording is not reported (clarification). (6) Independence recount: five Google pages = one party (the subject); The Decoder derivative ('Source: Google'); The Next Web an independent outlet but relays Google with limited added context; Artificial Analysis the only independent evaluator, and only for 1.1. Count of sources independent of Google: 2 outlets (TNW partial, AA) of which 1 holds independent measurements; no Analysis requirement applies (News). Anthropic disclosure: not required - the article does not mention Anthropic or Claude; the strings appear only in the Vertex docs navigation menu (Claude partner-model entries) and The Decoder's sidebar headlines, not in the article's subject, headline, summary or lead (Rule 15 check confirms the writer's statement).