Prototype · closed beta
unOpen.AI
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Model families
Gemini for
long-context work.
Gemini's distinguishing properties are context length and native multimodality — it was built to take video, audio, images and documents as first-class inputs rather than through a bolted-on pipeline. That combination decides a specific class of problem.
All Model Families →Introduction transcript. Qubi: Meetings, footage, recordings and documents rarely arrive as tidy text. This model takes video, sound and images together, so the timing stays intact. A long window isn't a memory, so we test recall at the length you'll use. Then we weigh the cost per call against retrieval, which can cite its source.
Current release
Gemini today.
Google ships these models as a range rather than one lead model, so this page is about the family: what it is good at, where it runs and how we evaluate it.
Last reviewed on 1 September 2026. We check the public model catalogues every week; when a new release appears, we review it against the provider's announcement and update this page.
LATEST MODELS
The newest Gemini models.
What Google has shipped most recently. Each has its own page with specifications, independent ratings at every reasoning setting and listed prices.
30 Sep 2026
Gemini 4 Argon
- Intelligence
- 50.4
- Context
- 1.05M
- In / out per M
- $2.00 / $10.00
2 Sep 2026
Gemini 3.8 Flash
- Intelligence
- 30.5
- Context
- 1.05M
- In / out per M
- $0.825 / $4.13
13 Aug 2026
Gemini 3.7 Flash
- Intelligence
- —
- Context
- 1.05M
- In / out per M
- $0.750 / $3.75
21 Jul 2026
Gemini 3.5 Flash Lite
- Intelligence
- 11.7
- Context
- 1.05M
- In / out per M
- $0.300 / $2.50
EVERY MODEL
All 15 Gemini models, compared.
Sort by any column. Select a model for its full page.
| Gemini 4 Argon | 30 Sep 2026 | 50.4 | 80 | 1.05M | $2.00 | $10.00 |
|---|---|---|---|---|---|---|
| Gemini 3.8 Flash | 2 Sep 2026 | 30.5 | 126 | 1.05M | $0.825 | $4.13 |
| Gemini 3.7 Flash | 13 Aug 2026 | — | — | 1.05M | $0.750 | $3.75 |
| Gemini 3.5 Flash Lite | 21 Jul 2026 | 11.7 | 399 | 1.05M | $0.300 | $2.50 |
| Gemini 3.6 Flash | 21 Jul 2026 | — | — | 1.05M | $0.750 | $3.75 |
| Gemini 3.5 Flash | 19 May 2026 | — | — | 1.05M | $1.50 | $9.00 |
| Gemini 3.1 Flash Lite | 3 Mar 2026 | — | — | 1.05M | $0.250 | $1.50 |
| Gemini 3.1 Flash Lite Preview | 3 Mar 2026 | — | — | 1.05M | $0.250 | $1.50 |
| Gemini 3.1 Pro Preview Custom Tools | 25 Feb 2026 | — | — | 1.05M | $2.00 | $12.00 |
| Gemini 3.1 Pro Preview | 19 Feb 2026 | — | 139 | 1.05M | $2.00 | $12.00 |
| Gemini 3 Flash Preview | 17 Dec 2025 | — | — | 1.05M | $0.500 | $3.00 |
| Gemini 2.5 Flash Lite | 17 Jun 2025 | — | — | 1.05M | $0.100 | $0.400 |
| Gemini 2.5 Flash | 20 May 2025 | — | — | 1.05M | $0.300 | $2.50 |
| Gemini 2.5 Pro | 5 Jun 2025 | — | — | 1.05M | $1.25 | $10.00 |
| Gemini 2.5 Pro Preview 06-05 | 5 Jun 2025 | — | — | 1.05M | $1.25 | $10.00 |
Intelligence and coding: Artificial Analysis indices at the model's best reasoning setting, and its measured output speed. Context and prices: the OpenRouter catalogue list price in USD per million tokens. Released: the provider's date where Artificial Analysis records one, otherwise the date the catalogue first listed it. Refreshed weekly; not measured by QQuantum.ai. Select a heading to sort.
Specification
- Provider
- Google, United States (EU regions available through Vertex AI)
- Weights
- Closed — API access only. The smaller Gemma family is separate and open-weight.
- Context window
- Among the longest available; documented at one million tokens or more on the top tier — check current documentation
- Modality
- Text, image, audio and video as native inputs
- Access paths
- Google AI Studio API, or Vertex AI inside your own Google Cloud project and region
- Where we use it
- Long-document and video/audio understanding, and where a client already runs on Google Cloud
Provider-published specifications as of 1 September 2026, not figures we measured. Model families move quickly — check the provider's own documentation before committing to a number.
Where it fits well
01
Genuinely long context
Where the alternative is aggressive chunking and reassembly, putting the whole corpus in the window removes an entire class of retrieval bug. It is not free — you pay for every token, every call — but for a one-off analysis over a large document set it is often cheaper than building the retrieval layer you would otherwise need.
02
Native multimodal input
Video and audio in, structured output back, without a transcription-then-analysis pipeline that loses timing and visual context between stages. For meeting analysis, media archives and inspection footage this is a structural rather than incremental advantage.
03
Vertex deployment
Consumed inside your own Google Cloud project and region, which frequently resolves data-residency and procurement questions using an infrastructure agreement you already hold. That path is often faster through security review than a new vendor relationship.
04
Model tiering
A wide capability and price range within one family, so routing between tiers does not mean integrating a second provider. The cheapest tiers are cheap enough to sit in high-volume paths where a frontier model could not.
The honest caveats
-
A long context window is not a memory.
Filling a million-token window and expecting uniform recall across it is the most common and most expensive mistake we see with this family. Retrieval quality degrades unevenly across a very long context in ways that are workload-specific, and the only way to know how yours behaves is to test it at the length you actually intend to use. A retrieval layer over the same corpus is frequently both cheaper and more accurate, and it gives you a citation. -
Cost scales with what you put in it.
Long-context calls are priced on input tokens, and a system that re-sends a large document on every turn of a conversation multiplies that by the conversation length. Caching a stable prefix helps materially, but only if the context is designed so the stable part is genuinely a prefix — that is an architectural decision worth making before the first line of integration code, not after the first bill. -
Weights are closed.
If your requirement is that no data leaves your hardware, this family is out and the conversation moves to open weights. Vertex in your own region resolves residency; it does not resolve "on our own metal".
INDEPENDENT RATINGS
How Gemini compares.
Third-party scores and listed prices, refreshed every week. Useful for building a shortlist; your own evaluation decides.
Composite index across their reasoning, knowledge, maths and coding evaluations · Higher is better · Linear scale from 0 to 66.0
- Gemini 4 Carbon
- 59.4
- Claude Opus 5.5
- 58.4
- GPT-6 Astra
- 56.9
- Gemini 4 Argon
- 50.4
- Muse Wonder Full
- 49.7
- Kimi K3.1 Pro
- 46.0
- Gemini 3.8 Flash
- 30.5
- Gemini 3.5 Flash-Lite
- 11.7
This family's highest-rated models, highlighted, against the best-rated model of each other family we cover. One composite score cannot show cost, latency, data residency or how a model handles your task.
Source: Artificial Analysis. Checked 30 October 2026. Measured independently by Artificial Analysis, not by QQuantum.ai or the providers.
USD per million output tokens · The newest releases in the family · Linear scale from 0 to $11.00
- Gemini 4 Argon
- $10.00
- Gemini 3.8 Flash
- $4.13
- Gemini 3.7 Flash
- $3.75
- Gemini 3.5 Flash Lite
- $2.50
- Gemini 3.6 Flash
- $3.75
Input is priced separately: Gemini 4 Argon $2.00, Gemini 3.8 Flash $0.83, Gemini 3.7 Flash $0.75, Gemini 3.5 Flash Lite $0.30, Gemini 3.6 Flash $0.75 per million tokens. Catalogue list prices; the provider's own rates, caching and batch discounts can differ.
Source: OpenRouter model catalogue. Checked 30 October 2026. Listed by OpenRouter, not measured by QQuantum.ai.
Questions
Questions about Gemini for Enterprise.
Should we use the long context instead of building retrieval?
For a one-off or low-volume analysis, often yes — it is dramatically less engineering. For a system that runs continuously over a corpus that changes, almost always no: you pay the full input cost on every call, you cannot cite a source span, and recall across very long contexts is uneven. The usual answer is retrieval for the system and long context for the exploratory work that designs it.
Does data sent to Vertex stay in our region?
Vertex offers regional endpoints and, at the time of writing, the data handling follows your existing agreement with Google Cloud — confirm against the provider's current terms — which is why it is often the faster procurement path. Confirm the specific region and the specific model availability, because not every model is served from every region, and that detail has derailed more than one deployment at the last stage.
Gemini or Claude or GPT?
For long-document and multimodal work Gemini has a structural advantage on paper; we still benchmark. For agentic tool use and instruction adherence under complex constraints we more often reach for Claude. For breadth of ecosystem and integrations, GPT. Anyone answering that confidently without seeing your workload is guessing, and we benchmark rather than assume.
