Google launched Gemini Decision 1 on November 11, a model that answers up to 1,000 yes/no, multiple-choice and rubric questions in one call without writing a single word. It bills $0.10 per million input tokens and nothing for output, the same rate as OpenAI’s Decisions API and about 2.4 times TypeSafe’s Jev at $0.042. Its real edge is audio, video and PDFs.
Key takeaways
- Gemini Decision 1 charges $0.10 per million input tokens and bills zero output tokens.
- One request can carry 1 to 1,000 questions inside a 262,144-token context window.
- On Google’s own seven-eval suite, Jev cost $0.68 to run versus $0.88 for Gemini.
What is Gemini Decision 1?
Gemini Decision 1 is a Google model that scores answers instead of generating them. You send one shared input, such as a support ticket, a product photo or a call recording, plus a list of closed questions. It returns a calibrated probability for each option in a single pass, so there is no text to parse and no output bill.
Google sells it as the Gemini Decisions API, and Decision 1 is the only model behind that endpoint. Sundar Pichai called it “built for closed-set choices,” according to The New Stack.
Google’s launch post on X put end-to-end latency at 150 to 250 milliseconds. That is fast enough to sit in front of every request an app handles, which is exactly where Google wants it.
The model handles three question types, per Google’s developer documentation: a yes/no check, a choice among 2 to 255 options, and a score on an ordered rubric. Every answer carries a confidence number between 0 and 1.
That confidence number is the product. Code can act automatically above a threshold and send everything else to a human reviewer or a bigger model.
How the Decisions API bills a request
The shared input is billed once per request, no matter how many questions ride on it. Each extra question adds only its own instruction and criteria tokens, Google’s documentation says, and latency stays nearly flat as questions are added.
In practice, checking a refund request against 40 policy rules costs little more than checking it against one. Classification work that used to need a full generative call, billed on output tokens, now bills on input alone.
How much does Gemini Decision 1 cost?
Gemini Decision 1 costs $0.10 per million input tokens on the paid tier and $0 for output, because it generates no output tokens. Google AI Studio also offers a free prototyping tier. That price is identical to OpenAI’s Decisions API and well above TypeSafe’s Jev, the startup decision model that launched in September.
| Model | Input per 1M tokens | Output | Inputs accepted |
|---|---|---|---|
| Gemini Decision 1 (Google) | $0.10 | $0 | Text, images, audio, video, PDF |
| Decisions API, GPT-6 Luna (OpenAI) | $0.10 | $0 | Text, images |
| Jev (TypeSafe AI) | $0.042 | $0 | Text |
| Gemini 3.5 Flash-Lite (generative) | $0.30 | $2.50 | Text, images, video, audio |
Google’s pricing page lists Decision 1 under its own heading and confirms the zero-output billing. OpenAI’s Decisions guide charges the same $0.10 for GPT-6 Luna, with no cache-read, cache-write or output charges.
TypeSafe prices Jev at $0.042 per million input tokens, with output “too cheap to meter,” InfoWorld reported at launch.
What does a billion tokens cost?
Scale makes the gap concrete. At list prices, one billion input tokens a month costs:
- $100 on Gemini Decision 1
- $100 on OpenAI’s Decisions API
- $42 on TypeSafe’s Jev
- $300 on Gemini 3.5 Flash-Lite, before any output tokens at $2.50 per million
Each vendor uses its own tokenizer, so the same document can count differently. Treat this as list-price math, not a bake-off.
Google’s free tier allows 200 requests per minute and 10,000 a day, and paid Tier 1 starts at 1,500 requests per minute, according to DataCamp’s review of Google’s limits. Free-tier data is used to improve Google’s products; paid-tier data is not.
Is Gemini Decision 1 better than Jev and OpenAI’s Decisions API?
On Google’s own benchmarks, mostly yes on accuracy and clearly yes on calibration, but not on price. Gemini Decision 1 leads 7 of the 9 accuracy tests Google published, while Jev wins the two tool-and-action tests and costs less to run across the shared suite. Every number below is Google-reported.
| Benchmark (Google-reported) | Gemini Decision 1 | Jev 1.14 | GPT-6 Luna | Gemini 3.5 Flash-Lite |
|---|---|---|---|---|
| JevBench (public split) | 92.2% | 90.0% | 90.9% | 86.6% |
| New Yorker Caption (text) | 79.7% | 72.0% | 68.4% | 73.7% |
| Mind2Web Multimodal | 86.2% | 86.8% | 81.8% | 78.0% |
| When2Call (tool calls) | 79.9% | 82.0% | 70.4% | 60.9% |
| Video-MME (no subtitles) | 69.1% | No score | No score | 63.3% |
| Top-1 calibration error (lower is better) | 0.0250 | 0.0494 | 0.0721 | 0.2104 |
| Cost to run 7 shared evals | $0.88 | $0.68 | $0.93 | $4.14 |
Calibration is the number Google leads with. An error of 0.0250 is about half of Jev’s 0.0494, so Gemini’s stated confidence tracks its real hit rate more closely.
For a business setting an auto-approve threshold, that matters more than a point of accuracy. A model that says 90% and is right 90% of the time lets you automate more without raising the error rate.
Where Jev still wins
Price is where the picture flips. Running the same seven shared evals cost $0.88 on Gemini Decision 1 and $0.68 on Jev, Google’s own chart shows. That makes Gemini about 29% more expensive for the same work.
Jev also beats it on When2Call, 82.0% to 79.9%. That test measures whether and how to call a tool, the exact decision an agent router makes most often.
Be skeptical of the headline efficiency claim, too. Google’s line that the API is nearly five times cheaper compares it with Google’s own Gemini 3.5 Flash-Lite ($0.88 versus $4.14), not with any rival.
Who wins and who loses from Gemini Decision 1?
Google wins distribution, OpenAI loses its head start, and TypeSafe loses some of the novelty that justified its valuation. The biggest winners are buyers of high-volume classification. Three serious vendors now sell the same zero-output product, at $0.042 to $0.10 per million input tokens.
- Google. Decision 1 works as a router in front of Google’s own generative models. Google says Gemini Notebook already uses it to pick the right model tier, cutting latency on everyday queries by up to 40%. Every routed call keeps the customer inside the Gemini API.
- OpenAI. Google matched OpenAI’s $0.10 rate to the cent and added audio, video and PDF input, which OpenAI’s endpoint does not accept. OpenAI’s docs say its API answers about 10 times faster than its Responses API, so speed alone no longer differentiates.
- TypeSafe AI. The startup raised $870 million at a $7.5 billion valuation 24 days after launching Jev. It keeps the lowest price and the tool-call lead, but now fights two of the largest labs with a text-only model.
- Google’s own Flash-Lite revenue. Classification jobs on Gemini 3.5 Flash-Lite paid $2.50 per million output tokens. Moving them to Decision 1 removes that line entirely. Google is cannibalizing its cheap tier before a rival does it for them.
There is a capital logic here too. Alphabet’s free cash flow hit -$9.1 billion last quarter while its cloud business doubled. A model that answers in one forward pass, rather than generating token by token, should be cheaper to serve per request, which is how a $0.10 price can still earn a margin.
What are the catches?
Three stand out. The API runs only on Google’s v1beta Developer API, it has no caching or batch discounts, and every benchmark so far comes from Google itself. Repeated inputs are billed in full each time, which can erase the price advantage for workloads that re-check the same long document.
- No context caching, Batch API, or Flex and Priority inference, per Google’s documentation.
- Not yet in Google Cloud’s enterprise platform; Google says that support is coming soon.
- Question instructions are text only, and the API rejects YouTube links and direct Cloud Storage URIs.
- Most of Gemini’s leads over Jev are one to two points, close to noise for a vendor-run test.
What to watch next
The next dated events are price changes elsewhere in Google’s lineup, which show how Google uses introductory rates to win volume. Beyond those, watch for Google Cloud availability and any price response from TypeSafe or OpenAI, since Decision 1 has no announced price clock of its own.
- December 1, 2026: Gemini 4 Carbon and Argon list prices double to $4 input and $20 output per million tokens, as we covered in our Gemini 4 Carbon pricing report. The same day, calls to the deprecated Gemini 2.5 Flash start routing to 3.5 Flash-Lite at Flash-Lite rates.
- January 1, 2027: Gemini 3.8 Flash’s input rate rises from $0.75 to $1.50 per million tokens, per Google’s pricing page.
- Signal: Google Cloud listing. Enterprise budgets run through cloud contracts. A Google Cloud launch would put Decision 1 inside existing commitments.
- Signal: a TypeSafe or OpenAI price move. Jev’s 2.4x price advantage is the startup’s main moat. A cut from OpenAI, or a multimodal Jev, would reset the table above.
FAQ
What is Gemini Decision 1?
It is the Google model behind the Gemini Decisions API. It answers closed yes/no, choice and rubric questions with calibrated probabilities instead of generated text. Google launched it on November 11, 2026.
How much does the Gemini Decisions API cost?
It costs $0.10 per million input tokens on the paid tier, with zero output tokens billed. A free prototyping tier is available in Google AI Studio.
Is Gemini Decision 1 cheaper than OpenAI’s Decisions API?
No. Both list $0.10 per million input tokens with free output. Gemini also accepts audio, video and PDF input, while OpenAI’s endpoint takes text and images.
Is Gemini Decision 1 cheaper than TypeSafe’s Jev?
No. Jev lists $0.042 per million input tokens. On Google’s own seven-eval suite, Jev cost $0.68 to run versus $0.88 for Gemini Decision 1.
Can Gemini Decision 1 write text?
No. It returns probabilities and confidence scores only. For summaries, extraction or tool calls, Google points developers to its generative Gemini models.
Where can developers use Gemini Decision 1?
Through the Gemini Developer API (v1beta), Google AI Studio and OpenRouter. Google says support in Google Cloud is coming soon.
The bottom line
Gemini Decision 1 is not a price cut. It is a price match with wider reach. Google charged exactly what OpenAI charges, stayed 2.4 times above TypeSafe, and won on the one thing neither rival offers: decisions over audio, video and PDFs.
For text-only routing, Jev remains the cheaper buy. For anything with media, Google just became the obvious first test. And decision models are now a three-vendor market, which is good news for whoever pays the bill.
Free guide · 12 pages
AI Tools Every Investor Should Use in 2026
10 tools for research, stock signals, charts and crypto, what each one costs, three ready-made stacks from $0 a month and 7 copy-paste prompts for filings and earnings calls. Enter your email and we will send it to you.
Plus one email a week: AI Money This Week, every Sunday. Unsubscribe anytime.
Sources
- Google AI for Developers: Decisions API documentation
- Google AI for Developers: Gemini Developer API pricing
- Google on X: Gemini Decisions API announcement
- OpenAI: Decisions API guide
- DataCamp: Gemini Decision 1 benchmarks, pricing and access
- The New Stack: Google built an AI model that can’t write a sentence
- InfoWorld: TypeSafe AI’s new models work with machines, not humans
Wealth Engine researches and drafts with AI tools and checks every figure against the sources above. How we report.
Leave a Reply