Models

CheapestInference serves six frontier models across three pools. A subscription is per-pool: during your reserved time blocks you get unlimited usage of every model in that pool — there is no separate full-catalog tier.

• Flagship Pool — Kimi K3, Qwen3.8 Max (the two flagship-class models, from $169.15/mo with annual billing — very limited seats)
• Frontier Pool — GLM 5.3, MiniMax M3 (frontier coding & agentic models, from $60.35/mo with annual billing)
• Core Pool — DeepSeek V4.1 Flash, MiMo v2.5 (fast, efficient models, from $15.29/mo with annual billing)

List models

Query the live, authoritative model list:

curl https://api.cheapestinference.com/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"

Each model object includes an id, owned_by, and a type field so you can filter programmatically:

{
  "id": "glm-5.3",
  "object": "model",
  "created": 1677610602,
  "owned_by": "cheapestinference",
  "type": "chat"
}

Flagship Pool

Model	Model ID	Max request size	Per-token price elsewhere (in / out per 1M)
Kimi K3	kimi-k3	8 MB	$3.00 / $15.00
Qwen3.8 Max	qwen3.8-max	8 MB	$2.00 / $6.00

From $169.15/mo with annual billing ($199/mo monthly). The two flagship-class models — Moonshot’s Kimi K3 and Alibaba’s Qwen3.8 Max (vision) — in one pool, very limited seats.

Frontier Pool

Model	Model ID	Max request size	Per-token price elsewhere (in / out per 1M)
GLM 5.3	glm-5.3	8 MB	$1.40 / $4.40
MiniMax M3	minimax-m3	8 MB	$0.30 / $1.20

From $60.35/mo with annual billing ($71/mo monthly). The Kimi K2 family (K2.6, K2.7) previously served here has been retired; for the Kimi family, Kimi K3 is served in the Flagship Pool.

Core Pool

Model	Model ID	Max request size	Per-token price elsewhere (in / out per 1M)
DeepSeek V4.1 Flash	deepseek-v4.1-flash	8 MB	$0.15 / $0.60
MiMo v2.5	mimo-v2.5	8 MB	$0.14 / $0.28

From $15.29/mo with annual billing ($17.99–21.99/mo monthly depending on the block).

Per-token prices are the models’ list prices elsewhere — reference only, useful for comparison. On a time-block subscription you pay a flat monthly fee, not per-token charges. See Plans & Limits.

The set of models in a pool can change over time, and more pools may open. GET /v1/models is always the authoritative live list.

Per-model details:

• Kimi K3 API — Moonshot’s flagship, served in the Flagship Pool
• Qwen3.8 Max API — Alibaba’s flagship with vision input, served in the Flagship Pool
• Kimi K2.7 — retired August 2026
• Kimi K2.6 — retired July 2026
• GLM 5.3 API — Zhipu’s coding & reasoning model
• MiniMax M3 API — frontier multimodal coding model
• DeepSeek V4.1 Flash API — DeepSeek’s fast, efficient model
• MiMo v2.5 API — Xiaomi’s fast, efficient model
• DeepSeek V4 Flash — retired September 2026, upgraded to DeepSeek V4.1 Flash
• Coming — models under review — the live pipeline of models being evaluated

Note

The model list above is a snapshot. The authoritative list always comes from GET /v1/models. Models may be added, swapped, or updated as new frontier releases ship.

Using models

Specify the model ID in your request:

# OpenAI SDK
response = client.chat.completions.create(
    model="glm-5.3",  # or "kimi-k3", "qwen3.8-max", "minimax-m3", "deepseek-v4.1-flash", ...
    messages=[{"role": "user", "content": "Hello"}]
)

All models work through the OpenAI endpoint (/v1/chat/completions) and the Anthropic-compatible endpoint (/anthropic/v1/messages). The API handles format translation automatically. Your key serves the models of the pool you subscribed to.

Request size limits

Each pool accepts requests up to a maximum payload size — messages, history and attachments combined: 8 MB per request on every pool (shown in the pool tables above). That is above the context window of every model served, so in practice the model’s own window is the limit you will meet first. Requests over the limit are rejected:

• Anthropic-dialect endpoints return {"type":"error","error":{"type":"invalid_request_error","message":"prompt is too long: <tokens> tokens > <limit> maximum"}}.
• OpenAI-dialect endpoints return error.code = "context_length_exceeded".

Agent clients such as Claude Code and OpenCode recognize these errors and compact the conversation automatically before retrying, so long sessions keep going. To have the client compact before a request is rejected rather than after, tell it the size to stay under: in OpenCode, add "limit": { "context": <context window>, "output": 32000 } to the model entry in opencode.json, using the model’s context window (262144 for Kimi K3, 1000000 for GLM 5.3, Qwen3.8 Max, MiniMax M3 and DeepSeek V4.1 Flash, 1048576 for MiMo v2.5) — see the OpenCode guide; in ZCode, set the model’s Context window to the same figure. The error message itself includes this tip for those clients.