AI model pricing & credits ranking
Every agent can run on a different model, and each model costs a different number of credits. This ranking groups the selectable models into three cost classes — $ Cheap, $$ Mid, $$$ Premium — so you can pick the cheapest model that still does the job. The exact charge is billed live from the AI Gateway's real per-token price; classes are the relative guide.
How model pricing works here
Credits are counted from each model's real per-token price at the AI Gateway: we meter the exact input, cached and output tokens a run uses and convert that cost to credits, so a model that costs its provider 50× more per token costs 50× more credits. What you pay is those credits × your plan's price per credit — that price covers the platform (hosting, orchestration, approvals, audit, support) as well as the model, which is why the dollar column below is higher than the provider's own list price.
So the relative cost between models is exact, and the absolute cost is credits × your plan's rate. You only pay premium-model credits on the agents you deliberately set to a premium model — keep them on a $ model and you pay $ credits.
For agent work, INPUT tokens are most of the cost, not output: measured on 337 production tasks, the median task read 32 input tokens for every output token, and output was under half of its model cost on all 49 models (about 29% on the median model). So a model's input and cache-read prices matter more than its output price — see credits by model. Cheaper models can be 20–30× less per token than premium ones. Because you set the model per agent, the simplest way to cut spend is to keep your specialist agents on a $ model and reserve premium models for the orchestrator (the planner) where quality matters most.
Want it in dollars? On Starter a credit costs $0.001 (100,000 credits = $100), so the ranking table below shows each model's credits per 1K tokens and the ≈ $ per 1M tokens a Starter customer pays — Scale and Agency pay less per credit. For what a whole task costs on each model, see credits by model; for each kind of task, credits by task type.
Cost classes
| Class | Approx. output cost | ≈ Price / 1M output | Best for |
|---|---|---|---|
| $ Cheap | ~4–8 credits / 1K output | $6.00 | Lowest credit cost — ideal for specialist agents. |
| $$ Mid | ~12–60 credits / 1K output | $36.00 | Balanced cost and capability. |
| $$$ Premium | ~120 credits / 1K output | $120.00 | Highest cost — reserve for the hardest work. |
Model cost ranking (cheapest first)
| Model | Provider | Class | Credits / 1K (in · out) | ≈ $ / 1M (in · out) | Reasoning |
|---|---|---|---|---|---|
| DeepSeek V4 Flash | DeepSeek | $ Cheap | 0.87 · 1.73 | $0.87 · $1.73 | — |
| Qwen3.5 Flash (cheap) | Alibaba (Qwen) | $ Cheap | 0.67 · 2.67 | $0.67 · $2.67 | — |
| GLM-4.7 Flash (cheap) | Zhipu (GLM) | $ Cheap | 0.47 · 2.67 | $0.47 · $2.67 | — |
| GPT-6 Luna (default) | OpenAI | $ Cheap | 0.67 · 3.33 | $0.67 · $3.33 | — |
| Grok 4.1 Fast (reasoning) | xAI | $ Cheap | 1.33 · 3.33 | $1.33 · $3.33 | Yes |
| Llama 3.3 70B (cheap) | Meta (Llama) | $ Cheap | 4.8 · 4.8 | $4.80 · $4.80 | — |
| GPT-6 Luna Fast (same model, 2× price, ~5s faster to a plan) | OpenAI | $ Cheap | 1.33 · 6.67 | $1.33 · $6.67 | — |
| GPT-5.6 Luna | OpenAI | $ Cheap | 1.33 · 8 | $1.33 · $8.00 | Yes |
| Gemini 3.1 Flash Lite | $ Cheap | 1.67 · 10 | $1.67 · $10.00 | — | |
| GPT-5 mini | OpenAI | $ Cheap | 1.67 · 13.33 | $1.67 · $13.33 | Yes |
| Gemini 2.5 Flash (cheap) | $ Cheap | 2 · 16.67 | $2.00 · $16.67 | — | |
| Gemini 3.8 Flash (default) | $ Cheap | 5 · 25 | $5.00 · $25.00 | — | |
| GPT-5.4 mini | OpenAI | $ Cheap | 5 · 30 | $5.00 · $30.00 | Yes |
| Claude Haiku 4.5 | Anthropic | $ Cheap | 6.67 · 33.33 | $6.67 · $33.33 | — |
| Gemini 3.5 Flash | $ Cheap | 10 · 60 | $10.00 · $60.00 | — | |
| Llama 4 Maverick | Meta (Llama) | $$ Mid | 1.6 · 6.47 | $1.60 · $6.47 | — |
| MiniMax M2.7 | MiniMax | $$ Mid | 2 · 8 | $2.00 · $8.00 | — |
| MiniMax M3 | MiniMax | $$ Mid | 2 · 8 | $2.00 · $8.00 | — |
| Mistral Large 3 | Mistral | $$ Mid | 3.33 · 10 | $3.33 · $10.00 | — |
| Qwen3.7 Plus | Alibaba (Qwen) | $$ Mid | 2.67 · 10.67 | $2.67 · $10.67 | — |
| DeepSeek V3.2 (cheap) | DeepSeek | $$ Mid | 4.13 · 12.33 | $4.13 · $12.33 | — |
| DeepSeek V3.2 Thinking (reasoning) | DeepSeek | $$ Mid | 4.13 · 12.33 | $4.13 · $12.33 | Yes |
| DeepSeek V4 Pro | DeepSeek | $$ Mid | 4.4 · 13.2 | $4.40 · $13.20 | — |
| Kimi K2 Thinking (reasoning) | Moonshot (Kimi) | $$ Mid | 3.13 · 13.33 | $3.13 · $13.33 | Yes |
| Grok 4.3 | xAI | $$ Mid | 8.33 · 16.67 | $8.33 · $16.67 | — |
| GLM-5.2 | Zhipu (GLM) | $$ Mid | 5.33 · 17 | $5.33 · $17.00 | — |
| Kimi K2.6 | Moonshot (Kimi) | $$ Mid | 6.33 · 26.67 | $6.33 · $26.67 | — |
| GLM-5.1 | Zhipu (GLM) | $$ Mid | 9.33 · 29.33 | $9.33 · $29.33 | — |
| Magistral Medium (reasoning) | Mistral | $$ Mid | 13.33 · 33.33 | $13.33 · $33.33 | Yes |
| DeepSeek R1 (reasoning) | DeepSeek | $$ Mid | 9 · 36 | $9.00 · $36.00 | Yes |
| Qwen3 Max Thinking (reasoning) | Alibaba (Qwen) | $$ Mid | 8 · 40 | $8.00 · $40.00 | Yes |
| Qwen3.7 Max | Alibaba (Qwen) | $$ Mid | 16.67 · 50 | $16.67 · $50.00 | — |
| Mistral Medium 3.5 | Mistral | $$ Mid | 10 · 50 | $10.00 · $50.00 | — |
| Claude Sonnet 5 | Anthropic | $$ Mid | 13.33 · 66.67 | $13.33 · $66.67 | — |
| Claude Sonnet 5.5 | Anthropic | $$ Mid | 13.33 · 66.67 | $13.33 · $66.67 | — |
| Gemini 2.5 Pro | $$ Mid | 8.33 · 66.67 | $8.33 · $66.67 | — | |
| GPT-5 | OpenAI | $$ Mid | 8.33 · 66.67 | $8.33 · $66.67 | Yes |
| GPT-6 Sol | OpenAI | $$ Mid | 13.33 · 66.67 | $13.33 · $66.67 | — |
| GPT-5.2 (cheaper) | OpenAI | $$ Mid | 11.67 · 93.33 | $11.67 · $93.33 | Yes |
| Claude Sonnet 4.6 | Anthropic | $$ Mid | 20 · 100 | $20.00 · $100.00 | — |
| GPT-5.4 | OpenAI | $$ Mid | 16.67 · 100 | $16.67 · $100.00 | Yes |
| Claude Opus 4.6 | Anthropic | $$ Mid | 33.33 · 166.67 | $33.33 · $166.67 | — |
| GPT-5.5 | OpenAI | $$ Mid | 33.33 · 200 | $33.33 · $200.00 | Yes |
| Gemini 3.1 Pro (preview) | $$$ Premium | 13.33 · 80 | $13.33 · $80.00 | — | |
| Claude Opus 4.7 | Anthropic | $$$ Premium | 33.33 · 166.67 | $33.33 · $166.67 | — |
| Claude Opus 4.8 | Anthropic | $$$ Premium | 33.33 · 166.67 | $33.33 · $166.67 | — |
| Claude Opus 5 | Anthropic | $$$ Premium | 33.33 · 166.67 | $33.33 · $166.67 | — |
| Claude Fable 5 | Anthropic | $$$ Premium | 66.67 · 333.33 | $66.67 · $333.33 | — |
| GPT-6 Astra (expensive) | OpenAI | $$$ Premium | 66.67 · 333.33 | $66.67 · $333.33 | — |
| GPT-5.5 Pro (expensive) | OpenAI | $$$ Premium | 200 · 1200 | $200.00 · $1200.00 | Yes |
How to choose
- Specialist agents (Analytics, Optimization, Creative, Research, Safety, Report) do well on a $ Cheap model — that's the default.
- Keep the orchestrator on a stronger model: plan quality drives the whole run.
- Reasoning models cost the most per token — reserve them for genuinely hard analysis.
- Running low on credits? Drop your specialists to a $ model to stretch the balance — see credits & billing and best practices.
Frequently asked questions
- Which AI model is cheapest?
- The $ Cheap class (e.g. Gemini Flash, DeepSeek, Llama 3.3 70B, GLM Flash) costs the fewest credits — typically ~4–8 credits per 1K output tokens, versus ~120 for premium models like GPT-5.5 Pro or Claude Opus.
- Does a cheaper model hurt quality?
- For executing the planner's steps (the specialist agents), cheap fast models are reliable — that's why they're the default. Plan quality matters most for the orchestrator, so keep that on a stronger model.
- How are credits charged per model?
- Per token: input + output tokens × the model's live AI Gateway price, converted to credits. A typical task runs ~1,350–1,900 credits; premium models cost much more per token than cheap ones — /docs/credits-by-model prices the same task on every model.
- How much is that in dollars?
- Credits × your plan's price per credit: $0.001 on Starter (100,000 credits = $100), less on Scale and Agency. The ranking table shows the ≈ $ / 1M a Starter customer pays for every model. Exact charges are billed live from the AI Gateway's per-token price and can differ.
- Is the credit price the same as the model provider's price?
- No. Credits are counted from the provider's real token price, so the RATIO between two models is exact; the price of a credit also pays for the platform — hosting, orchestration, the approval and audit layer, support — so a credit sells for more than the model cost it counts. Bring Your Own Model (paid plans) zero-rates model tokens if you would rather pay a provider directly.
- Can I set a different model per agent?
- Yes — every agent's model is switchable on the Agents page through the built-in AI Gateway, so you can mix a premium orchestrator with cheap specialists.