Models

This is the model-first index for ByteSpike. Pick the model you want; the per-model page tells you what it costs, what protocol it speaks, and how to make your first call.

​

Claude family

Anthropic’s reasoning-and-vision lineup, available via both the Anthropic Messages protocol and the OpenAI Chat Completions API. Cache control, tool use, and thinking blocks pass through transparently.

​

GPT family

OpenAI’s GPT lineup — Chat Completions for ubiquitous SDK code, Responses for Codex-style clients. Reasoning, structured outputs, web search, and vision all pass through.

​

Gemini family

Google’s Gemini lineup, exposed through the OpenAI Chat Completions API. The current flagship is gemini-3-1-pro (1M context); gemini-3-5-flash and gemini-3-flash cover the fast multimodal tiers.

​

DeepSeek family

The DeepSeek lineup via DeepSeek’s own HTTP API. V4 is dual-protocol (Chat Completions + Anthropic Messages); the older V3.2 is Chat-Completions-only. Vision not exposed on the HTTP API today.

​

字节跳动 / 豆包 (Doubao) family

ByteDance’s Doubao lineup — strong on Chinese-language tasks, vision-capable. Served over the OpenAI Chat Completions API.

​

Zhipu / 智谱 (GLM) family

Zhipu’s GLM lineup — Chinese-LLM general chat with vision support. Served over the OpenAI Chat Completions API.

​

Moonshot / Kimi family

Moonshot’s Kimi lineup — long-context Chinese-LLM. Served over the OpenAI Chat Completions API.

​

MiniMax family

MiniMax’s lineup — competitive Chinese-LLM general chat. Served over the OpenAI Chat Completions API.

​

Image and video

Image and video models live under their own protocol surfaces (/v1/images/generations for image, /v1/tasks/submit for async video). Per-model pages with current per-call rates live in /api-reference/image and /api-reference/video today; they’ll migrate into /models/ in a follow-up.

​

DOSIA recommended paths

If you’re calling ByteSpike from DOSIA, two protocol surfaces matter — and each one has a recommended primary model plus cheaper fallbacks. For DOSIA Cloud Enterprise admins building permission templates:

• Global edition preset → unlock claude-sonnet-4-6, claude-opus-4-8, gpt-5-4, gpt-5-4-mini, gpt-5-5, gemini-2-5-flash, gemini-3-1-pro. Default to Anthropic for Agent, OpenAI for Chat.
• China edition preset → unlock deepseek-v4-pro, deepseek-v4-flash, doubao-seed-2.0-pro, glm-5-1, kimi-k2-6, minimax-m2-7, deepseek-v3-2. Default to deepseek-v4-pro for Agent (anthropic-compat), doubao-seed-2.0-pro for Chat.

Both presets are admin-defined in the DOSIA Cloud Console; members never have to pick a model unless they want to override.

​

How to read a model page

Each page is the same shape:

1. Quickstart — a runnable request against the model via its primary protocol.
2. Capabilities — what the model itself supports (vision, tools, reasoning, JSON mode, streaming).
3. When to use — opinionated guidance.
4. Next — links to related models.

Same shape across families so you can scan side-by-side.

​

Next

• Endpoint types — the eight request shapes ByteSpike accepts
• Pricing — the per-model public rate card