Model Catalog

AIOrouter provides access to Chinese AI models through a single API key. Model availability is evidence-backed — only providers with verified credentials and live contract verification are available for routing.

Available Models

Availability Legend:

• ✅ Available — Credential obtained, adapter implemented, routable after live contract smoke passes
• ⏳ Pending — Business verification in progress; not yet routable
• ❌ Unavailable — Credential acquisition blocked (mainland China phone requirement); excluded from BETA

BETA Note (2026-05-12): DeepSeek, Qwen, and Kimi are launch candidates pending live contract smoke verification (P2-W7-019). GLM is pending business verification. Ernie and Doubao are excluded from BETA due to mainland China phone registration blockers. See docs/provider-credentials-registry.md and docs/provider-contract-verification.md for current evidence.

⚠️ Qwen, GLM, Ernie, and Doubao context windows are from internal estimates (2026-05-03). Not verified against live provider documentation. DeepSeek and Kimi data verified from official sources.

Model Selection Guide

DeepSeek V4 Pro — Flagship All-Rounder

• Strengths: General reasoning, code generation, multilingual (EN/CN/JP)
• Context: 1,000,000 tokens (1M)
• Max Output: 384,000 tokens
• Modes: Supports both thinking (CoT) and non-thinking modes
• Best for: Most tasks, coding, technical writing
• Source: api-docs.deepseek.com/quick_start/pricing

DeepSeek R2 — Chain-of-Thought Reasoning

• Strengths: Complex reasoning, math, step-by-step problem solving with visible CoT trace
• Context: 1,000,000 tokens (1M)
• Best for: Research, mathematical proofs, complex logic
• Note: DeepSeek's deepseek-chat / deepseek-reasoner model aliases are deprecated per official docs (2026-05). Thinking mode is now supported directly on the base models.

Qwen3-235B — Alibaba's Flagship

• Strengths: Multilingual, large context, DashScope-specific features (search augmentation)
• Context: Reportedly 128K — verify at help.aliyun.com
• Best for: Multilingual tasks, enterprise applications
• Note: Strict content moderation per Chinese regulations

GLM-5 — Bilingual Specialist

• Strengths: Strong bilingual (CN/EN) output, structured content generation
• Context: Reportedly 128K — verify at open.bigmodel.cn
• Best for: Enterprise reports, bilingual content, structured data extraction
• Note: Business verification pending

Kimi K2 — 1M Token Context Window 🏆

• Strengths: 1,000,000 token context window — excellent for long document processing and full codebase analysis
• Best for: Document summarization, research paper analysis, large-scale code review
• Note: The longest context in the catalog. 8× larger than most other models.
• Source: Moonshot API documentation

Ernie 5.0 — Enterprise-Grade

• Strengths: Chinese enterprise content, compliance documentation
• Context: Reportedly 128K — API access blocked, cannot verify live
• Best for: Chinese-language business materials, internal documentation
• Note: Very strict content moderation. API access blocked (requires China mainland phone).

Doubao Pro — Absolute Cost Leader

• Strengths: Lowest cost in the catalog
• Context: Reportedly 128K — contract tests pending
• Best for: Bulk processing, simple completions, cost-sensitive workloads
• Note: No function calling support. Most parameters are provisional pending contract tests.

Auto-Routing

If you don't specify a model, AIOrouter's intelligent router automatically selects the best provider based on:

1. Availability: Routes away from degraded providers
2. Capability: Matches model to request features (JSON mode, functions, etc.)
3. Context: Ensures request fits within the model's context window
4. Latency: Prefers providers with lower current latency

You can always see which provider handled your request in the X-Provider response header.

Pricing

Subscriptions include a provider-cost indexed monthly allowance. Token-equivalent usage varies by model cost: value models go further, while reasoning or western models consume allowance faster or may require prepaid credits.

See Billing Guide for subscription pricing and quota information.

Data freshness: DeepSeek data verified against api-docs.deepseek.com on 2026-05-06. Qwen, GLM, Ernie, and Doubao context windows have NOT been verified against live provider docs — they are from our internal model-capabilities.json (2026-05-03) and should be verified before public launch.