Gemini 3.5 Thinking
Chat model
Available on Ofrai — chat, images, and video in one place with shared credits.
Chat and video use credits from your Ofrai balance.
About 0.411408 credits per short prompt and 2.46845 credits per short reply (per ~1,000 characters of text).
Commercial use depends on the selected model and model terms.

About Gemini 3.5 Thinking on Ofrai
Gemini 3.5 Thinking is a reasoning-tier chat model in the Gemini lineup on Ofrai: it works through a problem before it answers, instead of replying in a single pass. It belongs to the 3.5 generation and is the entry point to explicit reasoning in this catalogue — the variant to reach for when a wrong answer is expensive and the question needs steps. On Ofrai it is billed from the shared credit balance like every other model. Expect a slower, heavier call than a Flash tier in exchange for a more considered answer.
About 0.411408 credits per short prompt and 2.46845 credits per short reply (per ~1,000 characters of text).
Gemini 3.5 Thinking model facts
- Model ID
- gemini-3.5-thinking
- Modality
- Chat model
- Released
- Not published
- Ofrai pricing
- About 0.411408 credits per short prompt and 2.46845 credits per short reply (per ~1,000 characters of text).
- Web search
- Not available
Best for
- Multi-step problems where the intermediate work matters: maths, logic, planning and constraint puzzles
- Debugging code or configuration where the cause must be traced before a fix is written
- Comparing options against stated criteria and showing how the recommendation was reached
- Drafting plans, specs and analyses that have to stay internally consistent from start to finish
Not the best fit when
- High-volume, price-sensitive traffic, where reasoning overhead makes a Flash tier the cheaper answer
- Latency-sensitive interactive flows that need a near-instant first token
- Straightforward lookups, rewrites and formatting that a Flash model already handles well
How it differs
Compared with Gemini 3.6 Flash, it trades latency and credits for reasoning depth: 3.6 Flash handles routine work faster and more cheaply, while 3.5 Thinking is the one to keep when the problem has to be solved rather than merely restated.
Example prompts
Work through this problem step by step, show each intermediate step and every assumption you had to make, then give the final answer on its own line at the end. [PROBLEM]
The code below fails with this error: [ERROR]. Diagnose the root cause before proposing anything, then give the corrected version and state plainly what was wrong. [PASTE CODE]
Evaluate options [A], [B] and [C] against these criteria in this order of weight: [CRITERIA]. For each criterion, show the trade-off, then recommend one option and name the condition under which you would switch to a different one.
Goal: [GOAL]. Constraints: [CONSTRAINTS]. Produce a step-by-step plan, then list the single biggest risk in it and how you would reduce that risk before starting.
Before answering, check the question below for ambiguity and hidden assumptions. If it cannot be answered as written, say exactly what information is missing. Question: [QUESTION]
FAQ
- What is Gemini 3.5 Thinking?
- Gemini 3.5 Thinking is the reasoning variant of Google's 3.5 generation on Ofrai. It works through a problem internally before producing an answer, which makes it stronger than a plain Flash model on multi-step analysis, maths and debugging. The trade-off is a slower call that draws more heavily on the shared credit balance, so it is a targeted tool rather than a default.
- When should I choose it over a Flash model?
- Choose it when the cost of a wrong answer exceeds the cost of the extra call: multi-step reasoning, maths, debugging, or options analysis with explicit criteria. Keep a Flash model for rewrites, summaries, classification and other work where the instruction is clear and the answer is short. A common setup is a Flash default with this model as the escalation route.
- Is it the newest reasoning model in the lineup?
- It is the reasoning-tier entry in this Gemini group, not the newest model overall. The 3.6, 3.7 and 3.8 Flash generations sit above it in recency, trading explicit step-by-step derivation for speed and cost. Pick 3.5 Thinking when the reasoning behaviour itself is what you need, and a newer Flash generation when latency and per-call cost dominate the decision.
- Does it cost more than a Flash model on Ofrai?
- Relatively, yes: a reasoning pass spends more tokens on a request than a single-pass Flash answer, so it draws more from the shared credit balance. Because that overhead is wasted on easy prompts, send it only the requests that need it and let a Flash tier absorb the routine majority. The platform publishes the exact per-model rate used for billing.
- Can it read images and documents?
- This entry is registered as a text chat model in Ofrai, so prompts are text. Google's Gemini family is itself broadly multimodal, but what a given platform exposes depends on that platform's integration, and reasoning tiers sometimes expose a narrower set of inputs than their Flash siblings. Treat multimodal support as something to verify per deployment rather than assume.
- How do I get reliable output from it?
- Put the constraints and the acceptance criteria inside the prompt, and ask for the reasoning to be shown separately from the conclusion. Long or ambiguous questions produce vaguer results, so instruct it to name any assumption it had to make. If you need strict formatting, specify the exact schema and verify the returned shape instead of trusting it.
Related models
More models in the same modality on Ofrai.
