Gemini Review 2026: Features, Pricing & Verdict

The Gemini that exists in 2026

Gemini is Google’s AI, and if that sentence sounds obvious, it’s worth noting how much has changed across the two years since launch. The early Gemini felt like a chatbot Google shipped because they had to. The 2026 Gemini is a coherent product that earns a place in serious users’ rotations — not because it’s the best at everything, but because it’s genuinely the best at a few specific things.

First, the market numbers: on 11 August 2026 Google said the Gemini app crossed one billion monthly active users — up from around 400 million a year earlier, and, Google says, its fastest-growing product ever. It’s a milestone that lands mid-distribution war, as OpenAI simultaneously turns its own free tier into an ad-supported product. One caveat worth keeping straight: that is a monthly active figure for the standalone Gemini app and web interface, and it excludes both AI Overviews in Search and Gemini features embedded in Workspace — so it isn’t directly comparable to the weekly active number OpenAI reports for ChatGPT. Either way, Gemini’s growth is driven mostly by Google shipping it directly into the products that 3 billion people already use — Search, Gmail, Docs, Android, YouTube. If you use any of those, you’ve probably used Gemini, whether you meant to or not.

That embedded-in-everything strategy is both Gemini’s biggest strength and its biggest identity problem. Is Gemini a product or a feature? Google hasn’t fully decided, and the experience reflects that ambiguity.

The model lineup (post-I/O 2026)

Google’s naming got cleaner in late 2025 and the May 19 I/O keynote extended the family further. As of late May 2026:

Gemini 3 Flash-Lite — cheapest, fastest, free tier lives here. Good for simple tasks, classification, quick lookups.

Gemini 3 Flash — still available as a default free-tier model with reduced quotas.

Gemini 3 Pro — the previous flagship; remains available. $2 per million input tokens (up to 200K context), $12 per million output. Above 200K context, input jumps to $4/M and output to $18/M. Paid-only since April 1, 2026.

Gemini 3.5 Flash (May 19, 2026) — surpassed Gemini 3 Pro on coding, agentic, and multimodal benchmarks. Superseded for most uses by 3.6 Flash below.

Gemini 3.8 Flash + Flash Cyber (NEW — September 2, 2026) — Google’s fourth Flash release in under four months and the current best everyday Flash. Takes text, images, audio and video; 1M-token context, 64K text output; 54.9% on HLE-Verified. Independent measurement (Artificial Analysis) puts it at Intelligence Index 59 on high reasoning, up 3 points from 3.7 Flash, at ~300 output tokens/sec and $0.58 cost per task on high. Pricing is $0.75/$3.75 introductory — identical to 3.7 Flash, and sharing the same Dec 31, 2026 expiry rather than getting its own window, so it reverts to $1.50/$7.50 on Jan 1, 2027 alongside every other Flash. Cached input is discounted 90%. Live in the Gemini app (AI Pro/Ultra), AI Mode in Search, Gemini in Sheets, and for developers via Antigravity, AI Studio, the Gemini API and Android Studio. Gemini 3.8 Flash Cyber shares the same foundation with more permissive cyber mitigations (86.2% CyberGym, 47.2% pass@1 CWE-Bench), gated behind the Fairwind programme for government bodies, critical-infrastructure operators and software maintainers — see the three-lab vetted-tier split.

Gemini 3.7 Flash (August 13, 2026) — Google’s “most intelligent workhorse model yet for coding and agents” superseded by 3.8 Flash above, three weeks after 3.6. Big first-party coding gains over 3.6 Flash: DeepSWE v1.1 49%→65.3%, FrontierCode 1.1 34.4%→43.6%, WebDev Arena Elo 1538→1588, AutomationBench 17%→30.4%. Introductory $0.75/$3.75 (half of 3.6 Flash) through Dec 31, 2026, then $1.50/$7.50 from Jan 1. Available in Gemini Spark (AI Pro/Ultra, 160+ countries), the Gemini API (AI Studio, Android Studio), Google Antigravity, and Gemini Enterprise. Benchmarks are Google’s own; no independent evals yet.

Gemini 3.6 Flash + Flash-Lite + Flash Cyber (July 21, 2026) — with 3.5 Pro still stuck, Google shipped three Flash models. 3.6 Flash ($1.50/$7.50) uses ~17% fewer output tokens than 3.5 Flash, jumps DeepSWE coding 37%→49%, lifts computer use to 83%, and advances the knowledge cutoff to March 2026 — superseded by 3.7 Flash above. 3.5 Flash-Lite ($0.30/$2.50) targets rock-bottom high-volume cost. 3.5 Flash Cyber is government/trusted-partner-gated. Google also teased Gemini 4.

Gemini 3.5 Pro (still unreleased — has now missed three deadlines) — the heavier flagship, first shown at I/O for a June launch, then early July, then July 17. All three passed with no release. Per reporting, Google scrapped a nearly-finished base model and restarted pretraining after it hallucinated too often, trailed GPT-5.6 on benchmarks, and broke down on recursive tool-calling and complex SVG generation. As of July 20 there is still no official date, no model card, no pricing, and no gemini-3.5-pro in the public API list. Its rumoured headline feature is a 2-million-token context window with a Deep Think reasoning mode — but Google has never confirmed those specs. A stopgap Flash release is reportedly in testing. Don’t plan around this model.

Gemini 2.5 Pro / Flash / Flash-Lite — still served, still undeprecated, and no longer adoptable. Since 18 September 2026 Google limits these three “to users who have actively used them in the past,” so an existing project runs indefinitely while a new API key or new Cloud project cannot call them at all. Google’s deprecation page lists no shutdown date for any of them and points new projects at 3.5 Flash-Lite or 3.8 Flash instead — both of which cost multiples more. If you have 2.5 in production, the migration guide prices every move and the two gotchas Google’s rate card hides. gemini-2.5-flash-image is the one family member with a published shutdown date: 15 March 2027, replaced by gemini-3.1-flash-lite-image.

Gemini Omni (NEW — May 19, 2026) — a series of multimodal models accepting image, audio, video, and text input and outputting editable video grounded in real-world knowledge. Available in the Gemini app, Google Flow, and YouTube Shorts. As of Google’s current deprecation table, gemini-omni-1.1-flash is also the named replacement for every Veo endpoint on the Gemini API, all of which shut down 22 October 2026 — so Omni did replace Veo for developers after all, whatever the I/O framing suggested. See the Veo review. AI Plus, AI Pro, and AI Ultra tiers only — Omni did not ship on the free tier despite pre-event speculation that Google would deploy the Nano-Banana-Pro distribution playbook for video.

Gemini Spark (NEW — May 19, 2026) — Google’s personal AI agent, Workspace-integrated and capable of taking actions across Gmail, Docs, Calendar, and other apps. Spark was announced as adopting MCP (Model Context Protocol), Anthropic’s agent-tool interoperability protocol, in summer 2026; Google has not published a shipped date for that integration. Availability (updated 7 September 2026): now Google AI Pro and Ultra subscribers — widened from Ultra-only at announcement — US only, English, Gemini mobile app. On 3 September 2026 Google began rolling out Google Photos control in Spark: editing and curating albums, creating shared collections, and turning photos into calendar events, gradually over several weeks.

Nano Banana Pro — Google’s image generation model, which got the memorable name and has become a serious photorealism competitor. More on this below.

Jules — Google’s asynchronous coding agent. Not a “model” exactly but a product tier that matters for developers.

The free-tier reality, as of 9 October 2026: the Gemini app serves free personal accounts Flash-Lite and nothing else. Flash and Pro both became paid-only on that date, and the Pro model is withdrawn from Google AI Plus too — on a date Google emails each subscriber rather than publishes. Omni is paid-only; Spark is paid-only at AI Pro or above. The API, AI Studio and Workspace are outside the change.

Pricing decisions that actually matter (post-I/O 2026)

The May 19 I/O keynote restructured Google’s AI subscription tiers materially. The new structure as of late May 2026:

Free tier — Flash-Lite only from 9 October 2026

(3.5) Flash-Lite with usage caps, and nothing above it. Until 9 October the free app also served (3.6) Flash and (3.1) Pro; both were withdrawn from unsubscribed personal accounts on that date. This is the second ratchet in six months — Pro went paid-only on 1 April 2026, Flash followed in October.

Treat the practical consequence as a change to how you evaluate Gemini, not just to what free users get: the free app is no longer a representative sample of the product, since the API has been on 3.8 Flash since 2 September 2026. Evaluate through AI Studio or the API instead, both unaffected. Full analysis of the change and the unpublished AI Plus cutoff date.

Google AI Plus — $4.99/month

Google’s entry paid tier, reduced from $7.99 in June 2026 and the cheapest paid tier among the big three. 2x access to Gemini against the free baseline, 400 GB of storage, and Flash-Lite plus Flash.

It no longer includes the Pro model. Google is removing Gemini 3.1 Pro from AI Plus on a per-account date delivered by email rather than a published one, which means a paying subscriber cannot look up when their plan shrinks. Read the version numbers before treating that as a large loss: the withdrawn model is 3.1 Pro, and Google’s own I/O 2026 keynote described 3.5 Flash as surpassing 3.1 Pro on benchmarks, with the app’s Flash since moved to 3.6. On capability, an AI Plus subscriber keeps the better model. What they lose is Deep Think — which was never on this tier — and the entitlement their plan page still advertises.

Buy it if: you want higher Gemini limits and 400 GB of storage cheaply. Skip it if: you specifically need Pro-class reasoning or Deep Think, which now start at AI Pro — a 4x price step.

Google AI Pro — $19.99/month

The direct competitor to ChatGPT Plus and Claude Pro. Unchanged in pricing post-I/O:

• Higher usage limits on Gemini 3 Pro and 3.5 Flash
• Gemini Omni access (video generation, AI Plus/Pro/Ultra-tier feature)
• Deep Search (web-research agent producing cited reports)
• Jules coding agent with 5x higher limits than free
• Gemini integration in Gmail, Docs, Sheets, Slides, and Meet with no usage limits
• Daily Brief (personalized digest of email, calendar, tasks — rolling out now)
• 5TB of Google Drive storage included, raised from 2TB on 1 April 2026 at no extra cost

The Drive storage bundling remains a structural advantage no competitor matches, and it got better in April 2026 when the included allowance went from 2TB to 5TB with no price change. If you were already paying for Google One 2TB ($9.99/month), upgrading to AI Pro for $10 more adds Gemini Pro and triples the storage.

Google AI Ultra (NEW) — $100/month

A new mid-tier introduced May 19. 5x Pro usage limits plus:

• Gemini Spark personal agent access (Workspace-integrated; US only — since widened to AI Pro as well, and extended to Google Photos on 3 September 2026)
• Higher Gemini 3 Pro / 3.5 Flash / Omni usage
• Priority queue across all features
• Storage starts at 20TB, against Pro’s 5TB

The $100 Ultra tier fills a gap between Pro ($20) and the heavy-user tier ($200) that previously didn’t exist. For users hitting Pro limits multiple times per week, this is the right size — comparable to Claude Max 5x ($100) in the Anthropic stack.

Google AI Ultra (heavy) — $200/month

The previous $249.99/month tier dropped to $200/month with the May 19 restructure. Same capabilities, lower price:

• Highest access to Gemini 3 Pro, 3.5 Flash, and Omni
• Jules with 20x higher limits (for multi-agent coding workflows)
• Highest Spark agent limits and priority access to new features

The $50/month price cut for existing $249 subscribers is a real consumer-friendly move. At $200/month, this tier now matches Claude Max 20x and ChatGPT Pro in pricing — Google removing the premium positioning that made the previous $249 tier feel out of step.

Usage model — compute-used (replaces daily prompt limits)

The May 19 keynote replaced the daily prompt-count cap with a compute-used model: prompts are scored by complexity, features used, and chat length, with a five-hour refresh cycle up to a weekly cap. Closer to OpenAI’s ChatGPT Plus model than the previous strict daily allotments. The change affects all tiers and is generally consumer-friendly — short prompts barely register against the cap, while heavy multi-step research sessions register proportionally.

API pricing

Gemini 3 Pro at $2/$12 per million tokens remains meaningfully cheaper than Claude Opus 4.7 ($5/$25) and competitive with GPT-5 on OpenAI’s API. Gemini 3.5 Flash API pricing was not detailed at the keynote — expected to land at the Flash tier (~$0.30/$1 per million tokens range based on Google’s previous Flash-tier pricing pattern). For developers building products on top of an AI model, the Gemini family remains the cost-per-quality winner in 2026.

Recommendation post-I/O:

• Free tier only for light, casual use — from 9 October 2026 it is Flash-Lite only, and no longer a fair trial of Gemini
• AI Plus at $4.99/month if you want higher limits and 400 GB of storage cheaply — but not for Pro-class reasoning, which left this tier in October
• AI Pro at $19.99/month for most paying users — the new 3.5 Flash + Omni + Drive storage bundle is excellent value
• AI Ultra at $100/month when you regularly bump into Pro limits OR want Spark agent access — this is the right new mid-tier
• AI Ultra at $200/month only for heavy Jules + Spark + Omni workloads where the 20x usage limits matter
• The previous $249 tier’s $50 price cut is automatic for existing subscribers

What Gemini genuinely does better

Research with citations. This is Gemini’s killer feature and the main reason serious users keep an AI Pro subscription alongside Claude and ChatGPT. Deep Search produces cited research reports that are genuinely more trustworthy than ChatGPT’s Deep Research — the citations link to real sources, the facts match those sources, and the synthesis is often sharp. For “what do I need to know about X topic?” work, Gemini is the default pick.

Google Workspace integration. If you use Gmail and Docs daily, Gemini is already there. Draft an email reply in Gmail with two clicks. Summarize a long thread. Generate a Doc outline inside Docs without switching tabs. Create a Slides presentation from a Doc. Build a formula in Sheets by describing what you want. For knowledge workers embedded in Google’s ecosystem, these integrations save more time than anything ChatGPT or Claude can offer — because there’s no context switch.

Handling Google’s data. If you ask Gemini “what’s on my calendar this week?” it actually knows. It can search your Gmail, read your recent Docs, reference your Drive files. This is either invaluable or creepy depending on your relationship with Google, but it’s genuinely useful for organizing your own digital life.

Image generation with Nano Banana Pro. Google’s image model is quietly excellent in 2026 — a serious photorealism competitor to DALL-E on text rendering and visual fidelity, though not quite as stylish as Midjourney for artistic work. Integrated directly in Gemini so you don’t leave the conversation. This is the image feature most people don’t know about.

Video generation with Gemini Omni (NEW — May 2026). The I/O keynote shipped Omni — multimodal video generation accepting image/audio/video/text input and outputting editable video. Available across AI Plus/Pro/Ultra tiers (not free). Competitive position vs Veo 3, Runway, and Kling is the most interesting development in the AI video category since Sora’s discontinuation. Early reports put Omni’s raw generation fidelity slightly behind ByteDance’s Seedance 2 but ahead on editing operations (object swapping, scene rewriting via chat instructions).

Gemini Spark personal agent (NEW — May 2026). Spark is Google’s answer to the agent-platform push from Anthropic and OpenAI. Workspace-integrated (Gmail, Docs, Calendar context), takes actions on behalf of the user, with MCP integration coming summer 2026. Currently US-only and AI Ultra tier ($100/month or above). For users who want an AI agent that actually handles multi-step work across their tools — not just chats — Spark is genuinely competitive with what Claude Spark agents do, and lands inside the Google ecosystem rather than requiring separate agent infrastructure.

Long-context handling. Gemini 3 Pro can work with 1M+ tokens in some configurations. Users routinely drop entire codebases and hour-long meeting transcripts in and get coherent responses. The context handling is comparable to Claude’s.

Free tier generosity. Even after the April 2026 changes, the free Flash tier is more generous than ChatGPT’s free tier and much more generous than Claude’s. For casual users who don’t want to pay, Gemini is the value pick.

Price-per-quality on API. For developers, Gemini 3 Pro is the cheapest frontier-capable model in 2026. This has real implications for startup economics.

Where Gemini falls short

Writing voice is still slightly off. Gemini has gotten much better at writing but it still has a faint “Google corporate” tone. Emails sound professionally polite. Blog posts sound informative but safe. Compared to Claude — which writes with genuine nuance — or ChatGPT’s better prompting responsiveness, Gemini’s default voice feels least human. You can work around this with prompting, but you’ll do more work.

Safety hedging can be excessive. Gemini declines or hedges on topics that ChatGPT and Claude answer directly. Asking for medical information, legal analysis, or anything with controversy gets you “I can’t give specific advice about…” more often than the competitors. For researchers this is a real friction.

Product sprawl. There’s Gemini, Gemini Advanced, Google AI Pro, Google AI Ultra, Google AI Plus, Gemini Code Assist, Gemini CLI, Jules, NotebookLM, Veo, Nano Banana Pro, Imagen, Deep Search. Google has a product for everything, which means figuring out which Google product to use for your task is a job itself.

Coding quality behind Claude. Gemini can code and Jules is a capable agent, but Claude Code (46% “most loved” in the JetBrains April 2026 survey) outclasses it on complex tasks. For simple snippet generation, Gemini is fine. For architectural work, Claude is better.

Inconsistent mobile experience. The Gemini mobile app is fine but feels less refined than ChatGPT’s. Voice input is available but not as smooth. The hand-off between “ask Gemini anything” and “Gemini inside an app” is sometimes confusing on phones.

Free-tier Pro removal hurt accessibility. The April 2026 change making Pro paid-only feels retrogressive. Previously, you could test the flagship model for free. Now you can’t. Competitive pressure will probably force Google to reverse this eventually.

Gemini vs. ChatGPT vs. Claude in 2026

A ranking by use case (see also the deep dive in Claude vs ChatGPT for the head-to-head):

For research with citations: Gemini > Perplexity > ChatGPT > Claude. Gemini’s Deep Search is the most trustworthy.

For working inside Google Docs/Gmail/Sheets: Gemini, with no real competition. The integration is the killer feature. For document-grounded research specifically, NotebookLM is Google’s purpose-built option.

For general questions: ChatGPT ≈ Gemini > Claude. Both are great at breadth.

For writing you’ll publish: Claude > ChatGPT > Gemini. Gemini’s voice issues are real.

For coding: Claude > ChatGPT > Gemini. Jules is capable but not best-in-class — see the best AI coding tools guide for how the leaders compare.

For image generation: ChatGPT ≈ Gemini (Nano Banana Pro) > Claude (can’t do it). Gemini’s image model is underrated.

For price-per-quality on API: Gemini > ChatGPT > Claude. Easily.

For broader category context, see best AI chatbots.

Who should use Gemini

• Anyone deep in Google Workspace — the integration is worth it alone
• Researchers who need cited sources — Deep Search is the best-in-class option
• Developers looking for a cheaper frontier model — Gemini 3 Pro API pricing is hard to beat
• Users who already pay for Google Drive storage — AI Pro nearly doubles as a storage upgrade
• People who want the best free tier — Flash is more generous than alternatives

Who shouldn’t use Gemini as primary AI

• Professional writers — Claude’s voice quality is meaningfully better
• People who want minimal guardrails — Gemini hedges more than competitors
• Users who need cohesive product design — Google’s AI product sprawl is confusing
• Engineers working on complex codebases — Claude Code remains the leader

The verdict (refreshed post-I/O 2026)

Gemini is the right pick for researchers, analysts, and anyone whose work involves synthesizing information from the web — and the May 19 I/O announcements materially strengthened the case across multiple axes. Gemini 3.5 Flash narrowing the gap to GPT-5.5 and Claude on capability while staying at Flash-tier cost; Omni adding video to the Gemini product surface; Spark answering the agent-platform push from Anthropic and OpenAI; and the pricing restructure (new $100/month AI Ultra mid-tier, $249 → $200 cut on the heavy tier) addressing the most-cited friction in Gemini’s prior pricing.

For Google Workspace users, Gemini was already the obvious pick, and the I/O updates made the bundle stronger — Daily Brief, Gmail Live (summer), Docs Live (summer), Universal Cart, plus Spark integrating across Gmail/Docs/Calendar. The 5TB Drive storage at AI Pro continues to make the math compelling for existing Google One subscribers.

What Gemini still doesn’t have is a clear lead on creative or deep writing work. Claude remains the leader for prose quality; ChatGPT for breadth and ecosystem. The “best AI for Google ecosystem users” positioning is now also “best AI for users who want a credible agent + video + research stack at consumer pricing.” That’s a broader market than the prior “research-only” framing.

The pragmatic call post-I/O:

• Google Workspace users: try Google AI Pro at $19.99/month — Workspace features + 3.5 Flash + Omni + Daily Brief easily clears $20 of monthly time saved
• Users hitting Pro limits regularly: the new $100 AI Ultra tier is the right intermediate step (5x Pro usage + Spark access). This tier didn’t exist before May 19 and fills a real gap
• Heavy users previously on the $249 tier: automatic $50/month savings as of the keynote
• Non-Google users wanting one AI subscription: Claude Pro or ChatGPT Plus remain better defaults; Gemini is the credible second subscription for research
• Developers building on AI APIs: Gemini 3.5 Flash at the new pricing is likely the cost-per-quality leader; worth running benchmarks against your specific workload

The May 19 I/O moved Gemini’s competitive positioning from “specialist research AI plus Workspace value” to “credible primary AI for users in or willing to adopt Google’s ecosystem.” That’s the meaningful shift. Refresh decisions accordingly.

Update, 28 August 2026: Gemini 3.5 Transcribe

Google put a dedicated speech-to-text model into public preview on 26 August 2026 — Gemini 3.5 Transcribe, available as gemini-3.5-transcribe for pre-recorded audio and gemini-3.5-transcribe-live for streaming. Google reports a 2.6% average word error rate non-streaming and 4.0% streaming as measured by Artificial Analysis, across 85+ languages with mid-session code-switching, plus a 70% improvement in time-to-final-transcription over the older Chirp 3 model.

It is already live inside Google products: Gboard Rambler on Android, the Gemini app for macOS, and Google Antigravity’s prompt box, with Chrome named as next.

For consumer Gemini subscribers this is invisible plumbing that should make dictation and transcription features quietly better. For anyone calling the API, the pricing is the interesting part — roughly $0.30 per audio-hour for batch with speaker diarization included, which meets the price floor the speech-to-text specialists already set rather than undercutting it, while repricing Google’s own Cloud Speech-to-Text by about 64%. Full analysis, including the session limits and the token-versus-per-minute billing catch, in our breakdown of the speech-to-text price floor.

Update, 6 September 2026: Lyria 3.5 music generation

Google brought Lyria 3.5, its music generation model, to the Gemini app on 4 September 2026 and opened it on the Gemini API, which entered public preview the day before. In the app it is available to all users globally — free accounts generate under rate caps with shorter maximum durations, while AI Pro and Ultra raise the caps and add stem export and faster queues.

For developers there are two model IDs: lyria-3.5 for full songs up to about three minutes at $0.08 per song, and lyria-3-clip-preview for 30-second clips at $0.04. Neither has a free tier. Output is 44.1 kHz stereo (MP3 by default, WAV available on 3.5), you can pass up to 10 images alongside the text prompt, and structure is controllable with Verse/Chorus/Bridge section tags. Generation is single-turn — there is no iterative editing across prompts — and named artist voices and copyrighted lyrics are refused. Every output carries a SynthID watermark applied at the model.

The per-song price is roughly 4x Suno Pro’s effective rate, so this is not a price attack on the category leader. What it is instead is the only frontier music model with a documented public API, which Suno still does not sell — the full argument is in our analysis of the launch.

Update, 17 September 2026: Google opened the smart home to rival agents

On 16 September 2026 Google opened early access to Home MCP, an official Model Context Protocol server that lets any MCP-capable agent read and control a Google Home. Documented clients are Google Antigravity, Claude and OpenClaw; launch coverage also names Hermes and ChatGPT. Five tools cover listing homes, enumerating devices and command schemas, reading live states, querying event history and running actions. Agents cannot unlock doors — that is refused server-side — and cannot create automations.

The detail worth registering for anyone weighing Gemini: Google built the open door before the proprietary one. The capability arrived for rival agents via an open protocol rather than as a Gemini feature, and nothing in the assistant experience described above changed on 16 September. It fits the pattern from Antigravity shipping with a rival’s model inside it — Google increasingly lets the model and assistant layer be contested while keeping the harness, the hardware and the subscription.

Requirements are steep for a consumer feature: Google Home Premium Advanced at $20/month, a Google Cloud project with the Home API enabled, a self-published OAuth application, and US English only.

What shipped in late September 2026

Two additions worth knowing about if you evaluate Gemini beyond the consumer app.

Gemini 3.8 Live with Live Avatar reached general availability in Gemini Enterprise on 24 September — real-time lip-synced video over Google’s speech-to-speech models, 97 languages with automatic detection, live camera and screen-share understanding, asynchronous tool calling, and US and EU endpoints. Both the audio and the video stream carry SynthID watermarks. The conversation bills on the Live API’s modality meter ($3 and $12 per million audio input and output tokens, about a penny a conversational minute); the avatar video has no published price at all, which is the part to plan for.

Gemini 3.8 Flash TTS and Flash-Lite TTS landed the day before with consent-gated voice replication — blocked outright in the EEA, UK and several other jurisdictions, and carrying a disclosed January 2026 price increase.

What changed

• 7 Oct 2026: Correction: AI Pro’s included Drive storage is 5TB, not 2TB — Google raised it on 1 April 2026 at no extra cost, and plan prices were re-verified unchanged at $4.99 / $19.99 / $99.99 / $199.99.
• 6 Oct 2026: Correction: gemini-2.5-flash-image has not shut down — Google lists it for 15 March 2027, replaced by gemini-3.1-flash-lite-image. The 3.1 Flash Image model it was previously said to migrate to is itself retiring 29 October 2026.
• 4 Oct 2026: Gemini 2.5 documented in the lineup for the first time — not deprecated, no shutdown date, but closed to new projects since 18 September; see the migration guide. Every Veo API endpoint retires 22 October with Gemini Omni as the replacement.
• 4 Oct 2026: The free Gemini app drops to Flash-Lite on 9 October and AI Plus loses the Pro model on an emailed, unpublished per-account date — analysis. Pricing section rewritten and the AI Plus tier documented for the first time.