How to call AICraft, how to choose which model answers, and what the router does when a channel misbehaves.
Everything on this page and on Pricing is quoted per million tokens, and tokens are not words. Two anchors:
Code and CJK text run denser than English — budget noticeably more tokens for the same visible length.
Grouped by model family, cheapest input first within each group. Representative rows only — the complete catalogue, with live rates, is on Models and Pricing.
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| DeepSeek-V4-Flash | 1.0M | $0.1601 | $0.3203 | Efficiency-tuned MoE (284B total, 13B active) with 1M context. |
| DeepSeek-V3.2-Exp-Thinking | 128K | $0.3203 | $0.4804 | Text model with 128K context. |
| DeepSeek-R1-0528 | 128K | $0.6406 | $2.5623 | Text model with 128K context. |
| DeepSeek-V4-Pro | 1.0M | $1.9217 | $3.8435 | Large MoE model (1.6T total, 49B active) with 1M context for advanced reasoning and coding. |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| Qwen3.7 Flash | 1.0M | $0.0320 | $0.1281 | Text model with 1.0M context. |
| Qwen3-Next-80B-A3B-Instruct | 131K | $0.1601 | $0.6406 | Text model with 131K context. |
| Qwen3.6-27B | 262K | $0.3690 | $3.6897 | 27B hybrid multimodal model with a 262K-token context window. |
| Qwen-2.5-VL-72B-Instruct | 128K | $2.5623 | $7.6870 | Text model with 128K context. |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| GLM-5.3-Flash | 1.0M | $0.1281 | $0.4484 | Text model with 1.0M context. |
| GLM-4.5 | 131K | $0.6406 | $2.5623 | Text model with 131K context. |
| GLM-4.6 | 200K | $1.1530 | $2.0178 | Text model with 200K context. |
| GLM-5.3 | 1.0M | $1.2812 | $4.4841 | Coding model with strong security skills: white-box code review and vulnerability discovery. |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| Doubao-Seed-1.6-Flash | 256K | $0.0240 | $0.2402 | Text model with 256K context. |
| Doubao-Seed-1.6 | 256K | $0.1281 | $0.3203 | Text model with 256K context. |
| Doubao-1.5-Vision-Pro | 128K | $0.4804 | $1.4413 | Text model with 128K context. |
| Doubao-Seed-2.1-pro | 256K | $0.9609 | $4.8044 | Production-grade model: upgraded coding, agent and multimodal abilities, with stronger… |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| GPT-5-Nano | 400K | $0.0553 | $0.4420 | Smallest, fastest GPT-5 variant for low-latency tools and real-time use, at the lowest cost. |
| GPT-5 | 400K | $1.3813 | $11.0500 | Reasoning model with test-time routing and reduced hallucination. |
| GPT-6-Sol | 1.1M | $2.2100 | $11.0500 | Text model with 1.1M context. |
| GPT-5.4-Pro | 1.0M | $33.1500 | $198.9000 | GPT-5.4 Pro adds deeper reasoning for high-stakes work: agentic coding and long-context… |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| Claude-Haiku-5.5 | 1.0M | $0.1105 | $0.5525 | Text model with 1.0M context. |
| Claude-4.5-Sonnet | 200K | $3.3150 | $16.5750 | Agent and coding model: system design, code security and long autonomous runs. |
| Claude-4.6-Opus | 1.0M | $5.5250 | $27.6250 | Opus model with 128K max output and adaptive thinking depth. |
| Claude-4.1-Opus | 200K | $16.5750 | $82.8750 | Text model with 200K context. |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| Gemini-2.5-Flash-Lite | 1.0M | $0.1105 | $0.4420 | Low-latency reasoning model for high-throughput, cost-sensitive use. |
| Gemini-3.5-Flash-Lite | 1.0M | $0.3315 | $2.7625 | Efficient model suited as a sub-agent for targeted tasks in multi-agent workflows. |
| Gemini-3.7-Flash | 1.0M | $0.8287 | $4.1437 | Better debugging and first-pass code accuracy than Gemini 3.6 Flash, at half the initial… |
| Gemini-3.1-Pro-Preview | 1.0M | $2.2100 | $13.2600 | Reasoning model for complex workflows: software engineering, agent reliability,… |
| Model | Context | Input $/M | Output $/M | In a word |
|---|---|---|---|---|
| Hy3-Preview | 262K | $0.1922 | $0.6406 | Efficient MoE for agent workflows with three reasoning levels and strong multi-step code… |
| Grok-4.1-Fast-Non-Reasoning | 2.0M | $0.2210 | $0.5525 | Multimodal model with a 2M-token context window, tuned for low cost per token. |
| MiniMax-M2.1 | 204K | $0.3363 | $1.3452 | Text model with 204K context. |
| Kimi-K2-Thinking | 256K | $0.6406 | $2.5623 | Text model with 256K context. |
| Grok-4.2-Non-Reasoning | - | $1.3813 | $2.7625 | Flagship model with strong agent tool calling and strict instruction following. |
| Kimi-K3 | 1.0M | $3.2029 | $16.0145 | Open-source 3T-class flagship with native vision and 1M context for long-horizon coding. |
| If you are… | Reach for | Why |
|---|---|---|
| Prototyping something | "model": "auto" | One string, no model choice to make, and fallback is handled for you |
| Shipping something reproducible | An explicit model ID | The ID passes through untouched, so the result is repeatable |
| Billing by volume | The cheapest input rate in the family you already trust | Input dominates the bill on long prompts |
| Writing long answers | Something with a low output rate | Output is usually the pricier half, and the half you generate |
| Whole-document work | A 1M+ window model | See the Context Window Guide |
Yes — if you call more than one provider and want one key, one bill, and one endpoint. No, if you only ever call a single provider directly.
Three tests, and you should be able to run all of them before you commit a production workload.
On AICraft, all three are checkable on public pages: the endpoint is in Docs, pass-through behaviour is stated above, and every rate is on Pricing.
Models, context windows and rates change often. Every number on this page is read from the live catalogue — Models for availability and context windows, Pricing for the billing table — and the live pages are the source of truth wherever this page and they disagree.
Context figures are the maximum input window published for each model. Rates are US dollars per million tokens. Only representative models are listed here; the full catalogue is on the Models page.