GPT-6 Luna
OpenAI · Quick tasks
- Est. credits
- 0
- Try
- 0
- Est. work time
- 0 min
Ready
Choosing an AI model
Give three models a simple job. Then give them a harder one. See when a cheaper model works and when it pays to switch.
Compare the same job Free demo. No sign-up or credits needed.
Try the email first. Then give the same models a harder job.
Pick up to three. A fourth pick replaces the first.
Credits are how you pay for AI in Studio. This free demo estimates how many each job takes.
OpenAI · Quick tasks
Ready
OpenAI · Everyday work
Ready
Anthropic · Hard problems
Ready
These jobs use sample results. Credit estimates include every try at each model’s listed rate.
Help me pick an AI
Tell us the job and what matters most. Get three models to start with.
Your shortlist · Fix or write code
Try one. Check it. Switch if you need to.DeepSeek
Coding and everyday automation on a tight budget.
Z.ai
Budget-friendly coding and analysis of text and images.
OpenAI
Focused edits, sorting information and frequent automations.
We suggest these models based on what their makers say they do well. Your plan sets which ones you can use. AI Studio and AI Hires share the credits in your plan. In Studio, you’ll see an estimate before you send.
See plans for this work03 / Follow the work
You do not pay for waiting. You pay for what the model reads and writes. Tap each step to see where your credits go.
Read 8,000Written 0example tokens so far. More tokens, more credits.
Step 01
Your instructions, files and chat history all count. The more it has to read, the more credits it uses.
Illustration, not a model speed comparison.
04 / Compare every model
Every dot is a model. Further right costs more. Tap up to three and compare them line by line.
One medium task. Credits for one example request: the model reads about 30,000 words and writes about 6,000.
The three lanes are our picks for where to start, not test scores. A higher lane does not always cost more.
| Claude Fable 5.2 | Gemini 4 Hydrogen | |
|---|---|---|
| Made by | Anthropic | |
| Good fit for | The hardest technical problems and open-ended projects. | Everyday coding and work involving several tools or steps. |
| Our laneA starting point, not a test score | Hard problems | Everyday work |
| One medium taskExample credits | 52 credits | 10 credits |
| ReadingCredits per 1M tokens | 640 | 120 |
| WritingCredits per 1M tokens | 3,200 | 600 |
| Can take in | Text, Images | Text, Images, Audio, Video, Files |
| ThinkingWorking a problem through before answering | Always on | Adjustable |
| Open weightsAnyone can download and run the model | Not listed | Not listed |
| ReleasedThe day its maker released it | Nov 9, 2026 | Nov 12, 2026 |
| Next step | Try it Maker's facts | Try it Maker's facts |
| Model | Made by | Our lane | Medium task | Reading | Writing | Can take in | Thinking | Open weights | Released by maker |
|---|---|---|---|---|---|---|---|---|---|
| DeepSeek Flash | DeepSeek | Quick tasks | 5 | 60 | 240 | Text, Images | Can be switched off | Not listed | Sep 10, 2026 |
| Gemini 3.5 Flash-Lite | Quick tasks | 7 | 60 | 500 | Text, Images, Audio, Video, Files | Adjustable | Not listed | Jul 21, 2026 | |
| GLM 5.3 Flash | Z.ai | Quick tasks | 2 | 30 | 100 | Text, Images, Video, Files | Always on | Yes | Aug 26, 2026 |
| Claude Haiku 5.5 | Anthropic | Quick tasks | 2 | 20 | 100 | Text, Images | Adjustable | Not listed | Oct 7, 2026 |
| Gemini 4 Hydrogen | Everyday work | 10 | 120 | 600 | Text, Images, Audio, Video, Files | Adjustable | Not listed | Nov 12, 2026 | |
| Gemini 4 Neon | Hard problems | 20 | 240 | 1,200 | Text, Images, Audio, Video, Files | Adjustable | Not listed | Nov 18, 2026 | |
| Gemini 4 Carbon | Hard problems | 13 | 160 | 800 | Text, Images, Audio, Video, Files | Adjustable | Not listed | Nov 10, 2026 | |
| GPT-6 Luna | OpenAI | Quick tasks | 2 | 20 | 100 | Text, Images | Can be switched off | Not listed | Sep 22, 2026 |
| GPT-6.1 Terra | OpenAI | Everyday work | 7 | 80 | 400 | Text, Images | Can be switched off | Not listed | Oct 26, 2026 |
| GPT-6.2 Sol | OpenAI | Everyday work | 13 | 160 | 800 | Text, Images | Always on | Not listed | Nov 17, 2026 |
| GPT-6.2 Astra | OpenAI | Hard problems | 52 | 640 | 3,200 | Text, Images | Always on | Not listed | Nov 17, 2026 |
| Kimi K3.1 Pro | Moonshot | Hard problems | 32 | 400 | 2,000 | Text, Images, Video | Always on | Announced | Oct 24, 2026 |
| Claude Sonnet 5.5 | Anthropic | Everyday work | 13 | 160 | 800 | Text, Images | Adjustable | Not listed | Sep 28, 2026 |
| Claude Opus 5.5 | Anthropic | Hard problems | 26 | 320 | 1,600 | Text, Images | Always on | Not listed | Sep 22, 2026 |
| Claude Fable 5.2 | Anthropic | Hard problems | 52 | 640 | 3,200 | Text, Images | Always on | Not listed | Nov 9, 2026 |
One thing they share: each maker lists a context window of about 1 million tokens, so how much a model can hold at once is not what sets these apart. Credit numbers are examples at today’s rates for short requests, with no tools and no saved-context discounts. Release dates are the day each maker released the model, not the day it arrived in cYpher.camp. Those dates and the facts about inputs, thinking and open weights come from each maker’s own pages, checked November 17–21, 2026. “Not listed” means the maker’s page does not say. What you can use in AI Studio may differ.
05 / Meet the whole lineup
Explore all 15 by what they can help you do.
Anthropic
The hardest technical problems and open-ended projects.
A starting point, based on what its maker says this model is good at. Your instructions, tools and thinking settings change the result too.
Try this model Read the provider source ↗What the makers’ own tests show
Anthropic reports that Fable 5.2 at High effort matched Opus 5.5’s best FrontierCode result for about half the cost per task.
Read the evaluation ↗In Anthropic’s speed test, Sonnet finished before Haiku. Haiku wrote faster, but Sonnet needed fewer tokens to get there.
Read the experiment ↗These are the makers’ own results under their own test conditions. They are not cYpher tests and they do not predict your job. Picks last reviewed November 12, 2026.
Your next step
Give a model one real task in AI Studio. Check its answer. When the job changes, compare your options again.
Tried a model? Found something useful? Compare notes with people in Camp →