Choosing a Model
Which Model should I choose on BanglaCodes?
There are three options: Auto (completely free on the free plan), Premium (Claude, GPT-5.5, Gemini, DeepSeek, which use Token) and your own API Key (zero cost). Auto is plenty for everyday site building and learning, so save Premium for bigger apps or stubborn bugs.
Three kinds of Model, three kinds of work
1. Auto: if you would rather not think about it, leave it on this. On a free account this is completely free. Once you buy Token and become Pro, a second Auto appears in the list, "Auto (best models)": it starts on even better (cheap premium) Model, and then Token gets spent; plain Auto still tries the free Model first, and only when they cannot, your Token, with a notice. There is a whole line of 15+ free AI standing ready: Gemini 3.7 Flash, Nemotron 3 Super, GPT-OSS 120B, Qwen3.8 27B, Nemotron 3 Ultra, GLM-4.7 Flash, and many more. If one is busy, the next one picks up the work by itself. You should not even notice. And on a day when no free Model can finish a build, that build runs on a premium Model (DeepSeek V4 Pro) at our expense, and if that fails too, on Gemini 3.7 Flash at our expense, up to 3 times a day (12 times in all per account), and the chat says so. With Token, pick "Auto (best models)" and every build runs on that Model.
2. Premium: hard work needs a big expert. Claude Fable 5.1, Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Haiku, GPT-6 Astra, GPT-5.5 Pro, GPT-5.5, GPT-5 Mini, Grok 4.6, Gemini 2.5 Pro, along with the powerful, low cost Chinese Model GLM-5.3 Flash, DeepSeek V4 Pro, GLM-5.3, Kimi K3, are all one click away in the "Premium" section at the top of the Model list. They eat far more Token (the rate is written in the list, for example GLM-5.3 Flash 40 Token/1K, DeepSeek V4 Pro 256 Token/1K, Haiku 400 Token/1K, Sonnet 1200 Token/1K), because there is a real cost to us behind them. With DeepSeek, GLM or Kimi you can get work of almost the same quality for far fewer Token. For a complicated app or a tangled bug there is nothing like them, but for everyday work Auto is plenty.
3. Your own Key: if you have an account with OpenAI, Anthropic, DeepSeek or xAI, put the Key into Settings. Then run that Model as much as you like, at zero cost.
New Model get added on their own
Every 6 hours we check the listings in two places, OpenRouter and OpenCode Zen, looking for new free Model. If the price is zero and it is good enough for the job, it joins the chain by itself. Which means when something new and good comes out, you get it automatically, without us having to do anything.
Some Model "think" a little
Some Model, like DeepSeek, think things through internally before answering. While that happens you will see "thinking deeply" on the screen. This is not it getting stuck, in fact these are the ones that answer hard questions best. The thinking itself is not stored in your chat, only the real answer comes through.
When it needs to understand a picture
If you attach a screenshot, that message goes only to Model that can read pictures (Gemini is in that group), not to the others. In Auto mode this sorts itself out, you do not have to think about it.
Which one, when?
- Everyday site building, learning, small tools: Auto is plenty.
- Big apps, complicated logic, work that keeps going wrong: try Claude or GPT-5 once and you will see the difference for yourself.
- Running Ollama on your own server? Add it under Custom Provider, that is free too.
Custom Provider (your own LLM)
In the "Custom Provider" section of Settings you can put in any OpenAI-compatible address. Ollama, LM Studio, vLLM, your office's own LLM, they all work.
You need three things: a name, a Base URL (for example https://llm.aiinbangla.com/v1) and the Model name (several, separated by commas). And the API Key if one is required, otherwise leave it blank.
One thing to keep in mind: the address has to be reachable from the internet. Writing http://localhost:11434 will not work, because that address means your own computer, and our server cannot reach it. If you want to connect the Ollama on your own computer, make an https link with Cloudflare Tunnel or ngrok and put that in.
Where to get an API Key
In the Provider list in Settings there is a "Get a Key" link beside each one that takes you straight to that site. Google (aistudio.google.com) and Groq (console.groq.com) both give Key for free, no card needed.
Key are stored encrypted with AES-256-GCM and are never shown in full, only the first and last few characters.
Beginner questions
I cannot tell which model to pick, must I?
No, leave it on Auto. It picks the best fit and falls to the next one by itself. Most people never touch it.
If one model is busy, is my work stuck?
No, the next free model tries automatically; above the message you will see "The best free AI is on it". Pick a model yourself (or use pro mode) and the one running is shown by name.
Is there truly a difference between premium and free results?
For simple work, barely. For large, complex apps premium (Claude, GPT-5.5) usually nails it in one pass. Try free first, upgrade only if needed.