The Short Answer
For most interviews, use a free Gemini key with Gemini 3.6 Flash. In our test it started answering in 1.2 to 3.5 seconds, it reads screenshots of coding questions, and it costs nothing within Google's free tier. Keep Gemini 3.1 Flash-Lite as the quick option: about 1 second to the first words.
For the hardest system design or reasoning questions, Gemini 4 Hydrogen is the smartest free choice, but it takes about 5 seconds to start. With a paid key, GPT-6.2 Sol and Claude Sonnet 5.5 are good everyday picks, and Claude Fable 5.2 or GPT-6.2 Astra are for the toughest questions, at about 4 times the price.
What Matters in a Live Interview
- Time to the first words. The interviewer is waiting. One to three seconds feels natural; ten seconds is a long silence. Ghotly AI shows the answer as it is written, so the first words matter more than the last.
- Screenshots. Coding and multiple-choice questions are read from your screen, so the model has to read images. Gemini, OpenAI, Anthropic and Grok models do; Groq and NVIDIA models mostly do not.
- Free limits. Free tiers limit how many requests you can make a minute and a day. Save a second free key (Groq or OpenRouter) as a backup: when one is used up, Ghotly AI answers with the other by itself.
- Cost. With a paid key you pay for every answer. The table below shows what one answer costs.
Speed: What We Measured
On 18 and 19 November 2026 we asked each free Gemini model the same interview question ("What is the difference between a process and a thread? Answer in 4 short bullet points.") from our server, with thinking switched off the way Ghotly AI asks. These are the times to the first words and to the whole answer. Your times also depend on your internet connection.
| Model | First words | Whole answer | Reads screenshots |
|---|---|---|---|
| Gemini 3.1 Flash-Lite | 0.9 to 1.5 s | 1.2 to 1.8 s | Yes |
| Gemini 3.5 Flash-Lite | 1.1 to 1.4 s | 1.4 to 1.5 s | Yes |
| Gemini 3.6 Flash | 1.2 to 3.5 s | 2.1 to 4.3 s | Yes |
| Gemini 3.8 Flash | 1.8 to 5.0 s | 2.8 to 5.9 s | Yes |
| Gemini 4 Hydrogen | about 4.9 s | about 5.3 s | Yes |
Google's servers are busier at some hours. In mid-November, Ghotly AI's error reports showed Gemini 3.6 Flash sometimes sending nothing for a whole minute in the Indian afternoon. Since version 4.3.2, the app asks Gemini 3.1 Flash-Lite with the same key when the chosen model has not started after 15 seconds.
We did not time the paid models, because we used no paid keys for this test.
Cost per Answer With a Paid Key
List prices per million tokens on 19 November 2026, from each provider's pricing page. In Ghotly AI an answer sends roughly 3,000 tokens (your resume, custom instructions, the last questions and the new one) and gets about 500 back. Reasoning models also bill their hidden thinking as output, so treat these as the least an answer costs.
| Model | Input / output per million tokens | One answer | An hour (30 answers) |
|---|---|---|---|
| Claude Haiku 5.5 | $0.10 / $0.50 | $0.0006 | $0.02 |
| GPT-6 Luna | $0.10 / $0.50 | $0.0006 | $0.02 |
| GPT-6.1 Terra | $1.00 / $5.00 | $0.006 | $0.17 |
| Gemini 4 Hydrogen (paid tier) | $1.50 / $7.50 | $0.008 | $0.25 |
| Grok 4.7 | $2.00 / $6.00 | $0.009 | $0.27 |
| GPT-6.2 Sol | $2.00 / $10.00 | $0.011 | $0.33 |
| Claude Sonnet 5.5 | $2.00 / $10.00 | $0.011 | $0.33 |
| Claude Opus 5.5 | $4.00 / $20.00 | $0.022 | $0.66 |
| GPT-6.2 Astra | $8.00 / $40.00 | $0.044 | $1.32 |
| Claude Fable 5.2 | $8.00 / $40.00 | $0.044 | $1.32 |
Prices change: check the provider's own pricing page before you buy credits.
Free Options
- Gemini (free tier): Gemini 3.6 Flash, Gemini 4 Hydrogen, Gemini 3.8 Flash, Gemini 3.5 Flash-Lite and Gemini 3.1 Flash-Lite. Gemini 4 Argon and Gemini 4 Carbon are paid only.
- Groq (free tier): GPT-OSS 120B and GPT-OSS 20B, very fast for spoken questions. For a screenshot, Ghotly AI tries Groq's preview vision model, but a Gemini key reads screenshots better.
- OpenRouter (free models): Auto, which picks only free models, and 14 free models, with a daily cap on free use.
- NVIDIA (free tier): Nemotron 3.5 Lightning, Mistral Large 4 and more, for spoken questions only in Ghotly AI.
Our Picks
| Situation | Pick | Why |
|---|---|---|
| Most interviews, free | Gemini 3.6 Flash | Fast, reads screenshots, free tier |
| You want the quickest start | Gemini 3.1 Flash-Lite, or GPT-OSS 20B on Groq | About 1 second |
| Hard design or reasoning questions, free | Gemini 4 Hydrogen | Google's newest, about 5 seconds to start |
| Paid, everyday | GPT-6.2 Sol or Claude Sonnet 5.5 | About 1 US cent an answer |
| Paid, cheapest | Claude Haiku 5.5 or GPT-6 Luna | Well under a cent an answer |
| Paid, hardest questions | Claude Fable 5.2 or GPT-6.2 Astra | Most capable, slowest, about 4.4 cents an answer |
How to Set It Up in Ghotly AI
- Get a free Gemini or Groq key (since 4.3.3 the same key also listens, so no Deepgram key is needed): our free API keys guide shows each step.
- In Ghotly AI, open Settings > API Keys, paste the key, press Test and pick the model.
- Add a second free key (Groq or OpenRouter) as a backup, so an answer still comes when one free limit is reached.
- Before the interview, try a practice question with a screenshot and with your voice. The how-to guide walks through the whole setup.
Common Questions
What is the best free AI model for interviews?
Gemini 3.6 Flash with a free Gemini key. In our test it started answering in 1.2 to 3.5 seconds and it reads screenshots. Gemini 3.1 Flash-Lite is even quicker to start, and Gemini 4 Hydrogen is smarter but slower.
Is Gemini 4 Hydrogen free?
Yes. On 19 November 2026, Google's pricing page listed Gemini 4 Hydrogen as free of charge on the free tier of the Gemini API, and $1.50 / $7.50 per million tokens on the paid tier. Gemini 4 Argon and Carbon are not available on the free tier.
Which AI model is fastest for interview answers?
Of the models we timed, Gemini 3.1 Flash-Lite and Gemini 3.5 Flash-Lite, at about 1 second to the first words. GPT-OSS 20B on Groq is also built for speed.
How much does an interview cost with a paid key?
About 30 answers in an hour cost roughly $0.02 with Claude Haiku 5.5 or GPT-6 Luna, $0.33 with GPT-6.2 Sol or Claude Sonnet 5.5, and $1.32 with Claude Fable 5.2 or GPT-6.2 Astra, before any thinking tokens.
Can I use ChatGPT Plus or Claude Pro instead of an API key?
No. Chat subscriptions work only in their own apps. Ghotly AI needs an API key from the provider, and several providers have free tiers.
