GEMINI 4 ARGON IS GENERALLY AVAILABLE -- ON THE API RATE CARD, IN THE GEMINI APP FOR PRO AND ULTRA, AND IN GITHUB COPILOT, AT AN INTRO PRICE THAT DOUBLES ON DECEMBER 1 (2026-10-13, vendor-primary): two weeks after the Fairwind-only announcement, Google posted 'Gemini 4 Argon is now available for everyone'. **Where:** 'globally available today for developers and enterprises via the API in AI Studio and Gemini platform, for Google AI Ultra subscribers in Google Antigravity, and Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search'; developers on the **Gemini API paid tier** in AI Studio and Android Studio; enterprises in Gemini for business, the API and Antigravity. The Gemini API changelog for 10/13 reads 'Gemini 4 Argon (gemini-4-argon) generally available (GA)'. **Price (ai.google.dev pricing page, fetched 10/15):** Standard input $2.00 per 1M 'through November 30, 2026' and $4.00 'starting December 1, 2026'; output (including thinking tokens) $10.00 then $20.00; context caching $0.10 then $0.20 with 'Implicit caching only (explicit caching is not supported)'; Free Tier 'Not available'; Batch/Flex $1 / $5 then $2 / $10; Priority $3.60 / $18 then $7.20 / $36. The launch post footnote matches: 'After the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply.' This answers the open question on the 9/30 entry below (the intro window now has an end date). **Claims (vendor-reported):** #1 on the Vals Index; ERQA 87% ('outperforms all other frontier models'); Gray Swan indirect prompt injection attack success cut to 0.7%; shortcut attempts on coding benchmarks down to 2%; external bio red-teaming unlocked under 1% of bio sub-tasks; RareBench 57% and LabBench 52.2% on Gamow Labs' private evals; Google says Argon in Antigravity raised agentic code submissions by more than 35% and cut rollbacks by nearly a third internally. **Also:** 10,000 scientists can apply for Antigravity's highest usage tier free for six months; the Fairwind Program expands to 'hundreds of additional' partners for guardrail-free cyber access; 'the rest of the Gemini 4 family' is promised with no date. GitHub Copilot added Argon the same day (see the GitHub Copilot page). No Gemini 4 Flash or Pro tier exists yet.
2026-10-13•Google (blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/, AI-vertical RSS pubDate Tue 13 Oct 2026; companion safety post gemini-4-model-safety-security/ same date) + ai.google.dev/gemini-api/docs/pricing (Gemini 4 Argon section) and /docs/changelog ('October 13, 2026') -- fetched 2026-10-15 via curl and Firecrawl
OCTOBER 6-8: 3.7 FLASH AND 3.5 FLASH DEPRECATED WITH AUTO-ROUTING, DEEP RESEARCH PRO PREVIEW SHUTS 10/23, NANO BANANA 2.1 GA, SYNTHID DETECTOR FOR EVERYONE, GEMINI AGENT FOR WORK (vendor-primary): **(1) Deprecations (Gemini API changelog, 10/08):** 'gemini-3.7-flash is deprecated and has been replaced by gemini-3.8-flash. It is no longer available to select in Google AI Studio, and API requests to gemini-3.7-flash are automatically routed to gemini-3.8-flash'; the same for **gemini-3.5-flash -> gemini-3.6-flash**. The deprecations table lists 3.7 Flash's shutdown as 'N/A' with the note that 'the endpoint is not shutting down'. Practical effect: pinned 3.7 Flash calls now bill and behave as 3.8 Flash. **deep-research-pro-preview-12-2025 shuts down 2026-10-23** (migrate to deep-research-preview-04-2026 or the -max variant). Lovable separately sets its own in-app retirement dates for 3.6 Flash (11/19) and 3.7 Flash (2027-01-28); those are not Google's. **(2) Nano Banana 2.1 GA (10/06)** replaces Nano Banana 2 on the API at $0.0336 per 1K image (see the nano-banana page). **(3) SynthID Detector global (10/07):** 'we're expanding access to everyone -- with the tool available globally in English starting today'; it checks whether an image, video or audio file was made with AI 'from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple'; Google says 180 billion+ images and videos and 240,000 years of audio have been watermarked and its built-in checks in Search, the Gemini app and Chrome handle over 1 million requests a day. **(4) Gemini agent (Google Cloud, 10/08, Gemini at Work 2026):** 'a universal agent for work' with an organization's business context that plans tasks, uses skills and tools, connects to business systems, chooses 'the best model for the job' and has built-in cost controls -- details are on the Google Cloud blog and were not independently verified here. **(5) Watch item as of 10/12:** the API models, pricing, changelog and deprecations pages fetched 10/12 contained no Gemini 4 or Argon entry (it went GA the next day, 10/13). **Also in window, not a Gemini change:** Google Labs launched Playground (playground.google, 10/07), an experimental prompt-to-game builder.
2026-10-08•Gemini API changelog ('October 8, 2026' and 'October 6, 2026' entries) and deprecations page (ai.google.dev, fetched 2026-10-12 via Firecrawl and WebFetch); blog.google 'Google expands SynthID Detector for AI content' (Oct 07, 2026) and 'Google Cloud launches Gemini agent' (Oct 08, 2026) -- fetched 2026-10-12 via curl from the blog.google RSS links
GEMINI 4 ARGON ANNOUNCED -- A FRONTIER MODEL YOU CANNOT USE YET: FAIRWIND-GATED ROLLOUT, A STATED INTRO PRICE, AND A 1M-TOKEN OUTPUT LIMIT (2026-09-30, vendor-primary): Google announced **Gemini 4 Argon**, 'our new frontier model, which is rolling out to a set of trusted cyber defenders through our Fairwind Program' -- the same gate it used for 3.8 Flash Cyber. **Access is phased:** 'We are actively engaged in the U.S. government's voluntary process for pre-release model access while we gradually expand access ... before making Argon available to developers, enterprises, and consumers as soon as possible.' **No date for general availability**, and the model is **not on ai.google.dev's rate card** (the pricing page was not re-fetched this sweep because the docs host redirect-loops curl; the launch post itself says 'will launch'). **Price, as stated by Google for launch:** 'Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price' -- the word 'introductory' matters, because both 3.7 Flash and 3.8 Flash carry intro rates that double on 2027-01-01; no expiry is given for Argon's. **Capability claims (all vendor-reported, no third-party verification):** output limit expanded to **1M tokens 'up from the previous 64K'**; **DeepSWE v1.1 77.9%** (new state of the art; compare GPT-6 Sol's 68.8% and Sonnet 5.5's vendor table); **AutomationBench 51.3%, ranked #1** (OpenAI's 9/22 table had GPT-6 Sol at 33.2%); leading on the Vals Index, Vals Finance Agent v2 and Harvey's Legal Agent Benchmark; **LVBench 91.7%** (long video); **CWE-bench v1 68%, tied first**, plus wins over 3.8 Flash Cyber on Google's internal vulnerability benchmark (20 languages) and Wiz's black-box pen-test benchmark; Gray Swan IPI prompt-injection robustness lead. Internal anecdotes: a 40% beat of a published quantum-resource baseline, 300 TiB of data-centre memory freed by Argon agents, C/C++ to Rust migrations up to the 800K-line Fuchsia Zircon kernel, and a libgav1 Rust decoder 2.7x faster than the prior port. **Cyber posture:** 'for trusted defenders and our own internal teams at Google, we'll be releasing Argon without cyber guardrails'; Wiz is using it through Scan for Good and found a critical healthcare-software exposure 'that previous frontier models had missed'. Safeguards before broad release: CBRN/cyber refusals with activation monitoring, prompt-injection hardening, **chain-of-thought and action monitoring that stops execution**, sealed sandboxes; Google 'strongly encourage[s] the rest of the industry to preserve reasoning transparency'. **What this meant for the page at the time:** the Gemini app, API and Workspace stayed on the 3.8 generation; Argon was a watch item. **Superseded 2026-10-13: Argon is now GA -- see the entry above.**
2026-09-30•Google (blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/, on-page date 'Sep 30, 2026', root RSS pubDate Wed 30 Sep 2026; footnote 1 on the introductory price) -- fetched 2026-10-02 via curl; ai.google.dev pricing not re-fetched (redirect loop), github.blog/changelog/ grepped for Argon with no hit
SKILLS REPLACE GEMS, WITH A DATED SHUTDOWN LADDER, AND GUIDED VISION LANDS IN GEMINI LIVE ON ANDROID (2026-09-30 and 2026-10-01, vendor-primary): **Skills (9/30):** 'rolling out skills directly into Gemini chat globally' for all Google AI subscription tiers (18+ now, under-18 'coming soon'), with Workspace business, enterprise, nonprofit and education 'in the coming weeks'. A skill is a saved set of instructions invoked by typing '/' and its name; Gemini can build skills from your chats and auto-run them on matching prompts; multiple skills stack; and 'starting today, you can create skills that include reference files like plain text documents, PDFs, or images'. Skills were already in Gemini Spark. **'Skills are replacing Gems':** sharing and Drive/Gemini Notebook attachments come 'in the coming weeks', and **Google 'will remove support for Gems starting in November for individuals with personal accounts, March 2027 for Workspace business, enterprise, and nonprofit customers, and June 2027 for Workspace education customers'**; Gems migrate into skills automatically; **Opal, the Labs mini-app builder, 'will also turn down at the same time in November'**, and Gems by Google Labs turn down without migrating. Three new dated flips. **Guided Vision (10/01):** 'launching Guided Vision in Gemini Live across compatible Android devices', built with the blind and low-vision community -- share the camera and Gemini gives audio descriptions plus reframing cues ('pan slowly to the right, tilt downward, or step back'); multilingual, tested with communities in India, Brazil, Singapore, Indonesia and Japan; enabled under profile settings > Use Guided Vision in Live, via an Android Accessibility shortcut or the TalkBack menu; **Android 9+ in regions and languages where Gemini Live is supported**. Google's own caveat: 'not a medical device, mobility aid, or white cane replacement' and 'not intended for navigation, safe-travel guidance, or obstacle detection'.
2026-10-01•Google (blog.google/products-and-platforms/products/gemini/automate-tasks-with-skills/, 'Sep 30, 2026', footnotes 1-2; blog.google/innovation-and-ai/products/gemini-app/guided-vision-gemini-live/, 'Oct 1, 2026') -- fetched 2026-10-02 via curl
GEMINI 3.8 FLASH TTS AND 3.8 FLASH-LITE TTS -- VOICE DESIGN FROM A PROMPT, VOICE REPLICATION FROM 30 SECONDS, LINE-BY-LINE DIRECTION, AND A DAY-ONE RATE CARD (2026-09-23, vendor-primary): Google introduced 'two new text-to-speech models to the Gemini family, transforming voice generation from static presets into a dynamic creative studio'. **Gemini 3.8 Flash TTS** -- 'built for deep creative direction and character design': generative voice design ('create bespoke voices from scratch by customizing role, accent and voice characteristics across more than 100 languages and dialects using natural language prompting'), a 2,000+ voice library with regional varieties (Mexican Spanish, Quebec French, Scots English), **voice replication from a 30-second sample** 'of your voice or a voice you have the rights to use' with consent verification (a verbal consent recording from the voice owner must match the reference speaker), SynthID watermarking and C2PA credentials, saved custom voices, voice remixing 'coming soon'. **Gemini 3.8 Flash-Lite TTS** -- 'built for high-volume, cost-efficient scale': dubbing, audio content, expressive voice agents. Both: line-by-line stage directions, long-form generation over hours with minimal speaker drift, native two-speaker scene staging from one script, scripted vocal bursts and backchanneling (<laughs>, <sigh>, <gasp>, |mhm|, |yeah|). **Benchmarks (Google-reported):** Flash TTS #1 overall on Hume AI's Voice Design Benchmark (71.4) and #1 in accent modeling (60.8); Flash TTS and Flash-Lite TTS #1 and #2 on Hume's Overall Quality Index; top positions on Voice Arena blind preference in Japanese, Brazilian Portuguese, Vietnamese, MSA, Mexican Spanish and Hindi. **Rate card (ai.google.dev pricing, fetched 9/28):** `gemini-3.8-flash-tts` **$0.50 per 1M text input / $9.00 per 1M audio output** ('equivalent to $0.00225 per 10s audio'), context caching $0.125; `gemini-3.8-flash-lite-tts` **$0.50 / $6.00** ($0.0015 per 10s); Batch tier half of each; all marked **'through December 31, 2026' and doubling 'starting January 1, 2027'** ($1.00/$18.00 and $1.00/$12.00) -- the same intro-then-double pattern as the 3.7/3.8 Flash text prices; free tier available for both. **Availability:** rolling out from 9/23 in the Gemini API and Google AI Studio (new audio playground with a dual-speaker screenplay editor and voice replication), 'coming soon via API in Gemini Enterprise'; for everyone, Flash TTS in Gemini Notebook and Flash-Lite TTS in Google Vids. Platform partners Agora, LiveKit, Pipecat, Vercel; integrators Figma, HeyGen, Linguana, Wondercraft, 99.co, Ollang. Competitive note: ElevenLabs' Eleven v4 (9/28) names 'Google Gemini 3.8 Flash-Lite TTS, and Google Gemini 3.8 Flash TTS' among the models it claims ~75% blind preference over -- the two vendors now benchmark directly against each other.
2026-09-23•Google (blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/, on-page 'Sep 23, 2026') + ai.google.dev/gemini-api/docs/pricing (Gemini 3.8 Flash TTS and 3.8 Flash-Lite TTS sections, via Firecrawl 2026-09-28) -- fetched 2026-09-28
GEMINI 3.8 LIVE WITH LIVE AVATAR -- A LIP-SYNCED VIDEO PERSONA ON THE LIVE MODEL, ENTERPRISE ONLY (2026-09-24, vendor-primary): nine days after Gemini 3.8 Live, Google introduced **Gemini 3.8 Live with Live Avatar**, 'bringing near real-time visual presence to our native live dialogue models. By pairing near real-time video generation with speech, the Live Avatar feature creates an experience that listens, sees, and speaks with a dynamic visual persona' -- precise lip-syncing, natural expressions, fluid turn-taking. Asynchronous tool calling keeps the dialogue running while tools fetch data (hotel check-in demo); native multilingual speech-to-speech synchronization across **97 languages** with lip-sync and expressions adapting mid-conversation. A library of preset avatars plus **custom avatars generated from a reference image 'while preserving reference likeness, brand styling, or character identity' -- custom creation 'currently available only through enterprise allowlisting'**. All audio and video output is SynthID-watermarked; a model card is published. **Availability: 'Starting today, Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise'** with API documentation -- not in the consumer Gemini app and no Gemini API price row found on the pricing page as of 9/28 (the 3.8 Live audio rates, $0.005/min in and $0.018/min out, are unchanged). This is the Google-side answer to avatar vendors (HeyGen, Synthesia) inside a native speech-to-speech model rather than as a post-processing layer.
2026-09-24•Google (blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-with-live-avatar/, on-page 'Sep 24, 2026') + ai.google.dev/gemini-api/docs/pricing (no Live Avatar row, 2026-09-28) -- fetched 2026-09-28
Issues here are sourced from our editorial sweeps, not real-time telemetry. Newer issues may exist.