Gemini API Pricing
Google DeepMind is the AI division of Alphabet, formed in April 2023 by merging the original DeepMind (founded London, 2010, acquired by Google in 2014) with Google Brain. CEO Demis Hassabis. Ships the Gemini family — the most multimodal frontier lineup on the market (text + image + audio + video native) — through both the Gemini API and the cloud platform now rebranded as Gemini Enterprise Agent Platform (formerly Vertex AI, renamed April 2026).
Gemini 3.1 Flash-Lite LAUNCHED MAY 2026
Standard text, image, and video input is $0.25/M; audio input is $0.50/M; output includes thinking tokens.
Gemini 2.5 Pro LAUNCHED JUN 2025
Tiered pricing: $1.25/M input, $0.125/M cached input, and $10/M output for prompts <= 200K tokens; $2.50/$0.25/$15 above 200K.
Gemini 2.5 Flash LAUNCHED JUN 2025
Standard text, image, and video input is $0.30/M; audio input is $1/M; output includes thinking tokens.
Gemini 2.5 Flash-Lite LAUNCHED JUL 2025
Standard text, image, and video input is $0.10/M; audio input is $0.30/M; output includes thinking tokens.
Gemini 3.1 Pro Preview LAUNCHED MAY 2026
Preview-stage Gemini 3.1 Pro.
Gemini 3.1 Flash-Lite Preview
Preview variant of Flash-Lite 3.1.
Gemini 3 Flash Preview LAUNCHED DEC 2025
Standard text, image, and video input is $0.50/M; audio input is $1/M; output includes thinking tokens.
Gemini 2.5 Flash-Lite Preview (09-2025)
Dated snapshot of 2.5 Flash-Lite from September 2025 preview drop.
Gemini 2.5 Computer Use Preview
Specialised preview for computer-use / browser-agent loops.
Gemini 2.0 Flash LAUNCHED FEB 2025
Google lists Gemini 2.0 Flash at $0.10/M input for text, image, and video, $0.70/M audio input, $0.40/M output, and $0.025/M cached text/image/video input.
Gemini 2.0 Flash-Lite LAUNCHED FEB 2025
Google lists Gemini 2.0 Flash-Lite at $0.075/M input and $0.30/M output.
Gemini 3 Pro Preview LAUNCHED NOV 2025
Archive price mirrors the Gemini 3 Pro Preview list rate preserved in the project snapshot: $2/M input, $0.20/M cached input, and $12/M output for prompts up to 200K tokens; tier 2 above 200K was $…
Gemini 1.5 Flash-8B LAUNCHED OCT 2024
Half-price variant of Gemini 1.5 Flash — at $0.0375 input this was the cheapest production-grade model on any first-party API in late 2024.
Gemini 1.5 Pro LAUNCHED MAY 2024
Archive price uses the last-published project snapshot: $1.25/M input and $5/M output, with a higher >128K prompt tier at $2.50/M input and $10/M output.
Gemini 1.5 Flash LAUNCHED MAY 2024
Archive price uses the last-published project snapshot: $0.075/M input and $0.30/M output, with a higher >128K prompt tier at $0.15/M input and $0.60/M output.
Gemini 1.0 Pro LAUNCHED DEC 2023
Google's first general-availability Gemini API model — released December 13, 2023 as part of the original Gemini 1.0 family (alongside Ultra and Nano).
| Model | Input /M | Output /M | Cached | Context | Max output | Vision | Tools | Tier |
|---|---|---|---|---|---|---|---|---|
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.03−90% | 1M | — | ✓ | ✓ | Light |
| Gemini 2.5 Pro | $1.25 | $10 | $0.13−90% | 2M | — | ✓ | ✓ | Frontier |
| Gemini 2.5 Flash | $0.3 | $2.50 | $0.03−90% | 1M | — | ✓ | ✓ | Mid |
| Gemini 2.5 Flash-Lite | $0.1 | $0.4 | $0.01−90% | 1M | — | ✓ | ✓ | Light |
| Gemini 3.1 Pro Preview FLAGSHIP | $2.00 | $12 | $0.2−90% | 1M | — | ✓ | ✓ | Preview |
| Gemini 3.1 Flash-Lite Preview PREVIEW | $0.25 | $1.50 | $0.03−90% | 1M | — | ✓ | ✓ | Preview |
| Gemini 3 Flash Preview PREVIEW | $0.5 | $3.00 | $0.05−90% | 1M | — | ✓ | ✓ | Preview |
| Gemini 2.5 Flash-Lite Preview (09-2025) PREVIEW | $0.1 | $0.4 | $0.01−90% | 1M | — | ✓ | ✓ | Preview |
| Gemini 2.5 Computer Use Preview PREVIEW | $1.25 | $10 | — | documented elsewhere | — | ✓ | ✓ | Preview |
| Gemini 2.0 Flash DEPRECATED | $0.1 | $0.4 | $0.03−75% | 1M | — | ✓ | ✓ | Deprecated |
| Gemini 2.0 Flash-Lite DEPRECATED | $0.07 | $0.3 | — | 1M | — | ✓ | ✓ | Deprecated |
| Gemini 3 Pro Preview RETIRED | — | — | — | Retired Mar 9, 2026 | → gemini-3-1-pro-preview | ✓ | ✗ | Retired |
| Gemini 1.5 Flash-8B RETIRED | — | — | — | Retired Sep 29, 2025 | → gemini-2-5-flash-lite | ✓ | ✗ | Retired |
| Gemini 1.5 Pro RETIRED | — | — | — | Retired Sep 29, 2025 | → gemini-2-5-pro | ✓ | ✗ | Retired |
| Gemini 1.5 Flash RETIRED | — | — | — | Retired Sep 29, 2025 | → gemini-2-5-flash | ✓ | ✗ | Retired |
| Gemini 1.0 Pro RETIRED | — | — | — | Retired Feb 18, 2025 | → gemini-2-5-flash | ✗ | ✗ | Retired |
The last 12 months at Google DeepMind.
Major releases, deprecations, and corporate moves. Sourced from ai.google.dev changelog, Google DeepMind blog, and recent press.
$0.10/$0.40aistudio.google.com and ai.google.dev — direct API access, free-tier playground, batch + flex tiers. Free quota for development; production runs on the paid pay-as-you-go rates shown here.
CLOUD · GCPRenamed from Vertex AI on April 23, 2026. Full agent stack — RAG, security command center, agent builder. Data residency at-rest available in US / EU / global multi-regions. Gemini 3.x is not yet in EU regions; 2.5 Pro and 2.0 Flash are available in europe-west4.
gemini.google.com — end-user product (free + Advanced subscription). Includes Workspace integration, Deep Research, image gen via Imagen, video gen via Veo. No API access.
PRODUCTIVITY · WORKSPACENative Gemini inside Docs, Sheets, Slides, Gmail, Meet. Enterprise plans get data-boundary guarantees and admin controls. Billed per seat, not per token.
Google DeepMind is the AI research and product division of Alphabet Inc. (the holding company spun out from Google in 2015). It was formed in April 2023 by merging the original DeepMind — founded in London in November 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman, acquired by Google for $400–650M in January 2014 — with the previously separate Google Brain team inside Google. CEO: Demis Hassabis.
The merger ended a decade of internal duality. Pre-merger, Google ran two distinct AI orgs: Google Brain (Mountain View, headed by Jeff Dean) shipped TensorFlow, the original Transformer paper, and most of Google's product AI; DeepMind (London, headed by Hassabis) produced AlphaGo, AlphaFold, and most of the lab's public scientific breakthroughs. The merge consolidated leadership and unblocked the shipping cadence that produced the Gemini family.
Major technical contributions: AlphaGo / AlphaZero / MuZero (game-playing systems that beat world champions at Go, chess, shogi); AlphaFold (protein structure prediction — 2024 Nobel Prize in Chemistry for Hassabis and John Jumper); the original Transformer architecture (Vaswani et al., 2017, Google Brain); Gemini (multimodal LLM family, first released Dec 6, 2023); Imagen (image gen), Veo (video gen), and Lyria (music gen).
As an Alphabet subsidiary, Google DeepMind benefits from Alphabet's full revenue base (~$350B annual, 2025) and Google Cloud's compute footprint. It does not raise external venture funding the way Anthropic and OpenAI do — capital and TPU access come from inside the org. The catch: AI products compete for runway against Google Search, Google Cloud, and YouTube within the same P&L; product velocity has historically been slower than smaller pure-play labs.
Compared to OpenAI and Anthropic: the deepest multimodal stack (text + image + audio + video natively in one model), the longest context window on a flagship (2M on Gemini 2.5 Pro), and the broadest distribution via Workspace + Gemini app + Android. Weaker on coding-agent benchmarks (Anthropic still leads SWE-bench Verified; Gemini 3 Flash hits 78% vs Claude Opus 4.7's 87.6%) and on consumer brand recognition for ChatGPT-style chat.
OpenAI
Bigger consumer brand via ChatGPT, deeper Microsoft / Azure integration, faster product cadence on the API. Behind Google on max context window (1M vs 2M) and on native video / audio modalities.
CODING SPECIALISTAnthropic
Leads SWE-bench Verified at 87.6% with Claude Opus 4.7 — Google's best is Gemini 3 Flash at 78%. Stronger on zero-retention defaults and Responsible Scaling Policy. Smaller multimodal surface.
WILDCARDxAI
X / Twitter native integration and Memphis Colossus supercluster. Smaller multimodal surface and shorter context windows than Gemini; ecosystem is the moat or the bottleneck depending on your stack.
EU CHAMPIONMistral
EU sovereignty and GDPR-native by construction. Stronger French / German support. Lower frontier benchmarks but more transparent provenance; viable on EU compliance contracts where Vertex still lacks Gemini 3 in EU regions.
Meta
Llama 4 405B self-hostable — full weights, Apache-style licensing, deployable on your own hardware. Google's open-weight answer is Gemma, smaller-scale models tuned for on-device.
Frequently asked.
Practical questions about Google DeepMind, the Gemini lineup, and Vertex / Gemini Enterprise Agent Platform.
Q · 01 Which Gemini model should I start with? +
2M), and is cheaper than the 3.1 preview at $1.25/$10. Move to Gemini 3.1 Pro Preview for newest reasoning + the 3.x preview surface. Move down to Gemini 2.5 Flash for daily traffic, 2.5 Flash-Lite for classification. For coding agents, Gemini 3 Flash Preview hits 78% SWE-bench Verified at a quarter the cost of Pro.Q · 02 How does Gemini compare to Claude and GPT? +
Q · 03 Does Google offer EU data residency? +
europe-west4 (Netherlands) for GDPR-compliant workloads. Gemini 3.x is not yet in EU regions; if you need the newest models, you're still pinned to US / global endpoints, which the Vertex docs warn do not guarantee in-region processing.Q · 04 What's the data retention policy? +
Q · 05 When will Gemini 3 hit general availability? +
May 19, 2026. Cadence so far: Gemini 3 Pro Preview shipped Nov 2025, retired Mar 2026 in favour of 3.1 Pro Preview; Gemini 3 Flash Preview shipped Dec 2025; Gemini 3.1 Flash-Lite hit GA on May 7, 2026. GA for 3.1 Pro is expected later in 2026 once the preview surface stabilises — Google has not announced a date.Q · 06 Can I get a discount? +
Q · 07 What about Gemini 1.5 / 2.0? +
Sep 29, 2025. Gemini 2.0 Flash + 2.0 Flash-Lite are deprecated as of Feb 18, 2026 with full shutdown scheduled for Jun 1, 2026 — migrate to 2.5 Flash / 2.5 Flash-Lite. Both old-family slugs still take traffic until the retirement date but are not recommended for new builds.