AI Model Lineup

Offering diverse AI model options to meet different scenarios and performance needs

Service transparencyService boundaries and risk notices are shown
No multipliers · official pricingNo rate multipliers — billed exactly at official API prices
Logs queryable & exportableEvery request log can be queried and exported in the console

Some model prices reference the model provider public pricing page; final costs follow console display and provider rules

Standard Series

gpt-5.6-sol

Flagship of the GPT-5.6 family and OpenAI's strongest reasoning and agentic model to date. Scores 42.2 on Agents' Last Exam — a benchmark of long-horizon workflows across 55 domains — leading peer models by a wide margin at lower cost. Excels at complex coding, long-range planning, and deep research, with up to 128K output tokens for demanding engineering and automation tasks.

Supported Depths:LowMediumHighxhigh
Input$5.00/M·Cached Input$0.50/M·Output$30.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

gpt-5.6-terra

The balanced model of the GPT-5.6 family, matching GPT-5.5-level performance at half the price. Strikes the best balance between capability, speed, and cost — ideal as the workhorse for everyday coding assistance, production workloads, and internal tools.

Supported Depths:LowMediumHighxhigh
Input$2.50/M·Cached Input$0.25/M·Output$15.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

gpt-5.6-luna

The cost-efficient model of the GPT-5.6 family, delivering strong capability at OpenAI's lowest price tier. Fast responses and high throughput make it the go-to choice for latency-sensitive or high-frequency workloads such as summarization and classification.

Supported Depths:LowMediumHigh
Input$1.00/M·Cached Input$0.10/M·Output$6.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

grok-4.5

xAI's flagship model, trained alongside Cursor and built for coding, agentic tasks, and knowledge work. Features a 500K-token context window and configurable reasoning depth (Low/Medium/High), with roughly 2× the token efficiency of comparable models — completing tasks in fewer steps with markedly fewer hallucinations.

Supported Depths:LowMediumHigh
Input$2.00/M·Cached Input$0.50/M·Output$6.00/M
Model providerxAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

gpt-5.5

An enhanced step up from GPT-5.4 at 2× the per-token cost, retaining the same 1M-token context window. Improves 9 of 10 shared benchmarks — ARC-AGI-2 (+11.7pp), MCP Atlas (+8.1pp), and Terminal-Bench 2.0 (+7.6pp) see the largest gains. Completes identical Codex tasks with fewer tokens and fewer retries, ideal for high-precision complex engineering.

Supported Depths:LowMediumHighxhigh
≤272KInput$5.00/M·Cached Input$0.50/M·Output$30.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

gpt-5.4-mini

Lightweight and fast version of GPT-5.4, delivering the core capabilities of the GPT-5 series at a highly competitive price. Faster responses and lower cost make it ideal for latency-sensitive or high-frequency use cases like everyday coding assistance and conversational tasks.

Supported Depths:LowMediumHigh
Input$0.75/M·Cached Input$0.075/M·Output$4.50/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

gpt-5.4

The flagship new version of the GPT-5 series, achieving comprehensive breakthroughs in multimodal understanding, long-form reasoning, and complex task planning. Supports four reasoning depths — Low, Medium, High, and xhigh — covering all scenarios from everyday conversation to advanced research.

Supported Depths:LowMediumHighxhigh
≤272KInput$2.50/M·Cached Input$0.25/M·Output$15.00/M
>272KInput$5.00/M·Cached Input$0.50/M·Output$22.50/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

Claude Code Series

Use these models inside the Claude Code client through protocol format adaptation. See thesetup guide

cc-gpt-5.6-sol

Claude Code-compatible mapping of GPT-5.6 Sol. Via Codex → Claude Code protocol conversion, you can call it directly inside the Claude Code client and enjoy GPT-5.6 Sol's flagship reasoning and agentic capabilities at the same pricing as gpt-5.6-sol.

Supported Depths:LowMediumHighxhigh
Input$5.00/M·Cached Input$0.50/M·Output$30.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

cc-gpt-5.6-terra

Claude Code-compatible mapping of GPT-5.6 Terra. Via Codex → Claude Code protocol conversion, you can call it directly inside the Claude Code client and enjoy GPT-5.6 Terra's balanced performance at the same pricing as gpt-5.6-terra.

Supported Depths:LowMediumHighxhigh
Input$2.50/M·Cached Input$0.25/M·Output$15.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

cc-gpt-5.5

Claude Code-compatible mapping of GPT-5.5. Via Codex → Claude Code protocol conversion, you can call it directly inside the Claude Code client and enjoy the high-precision capabilities of GPT-5.5 at the same pricing.

Supported Depths:LowMediumHighxhigh
≤272KInput$5.00/M·Cached Input$0.50/M·Output$30.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

cc-gpt-5.4

Claude Code-compatible mapping of GPT-5.4. Via Codex → Claude Code protocol conversion, you can call it directly inside the Claude Code client and enjoy the full capabilities of GPT-5.4 at the same pricing.

Supported Depths:LowMediumHighxhigh
≤272KInput$2.50/M·Cached Input$0.25/M·Output$15.00/M
>272KInput$5.00/M·Cached Input$0.50/M·Output$22.50/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

Featured Series

codex-auto-review is for Codex automatic approval review flows, not a daily chat model. See thesetup guide

codex-auto-review

A specialized model for Codex remote automated review. OpenAI Codex switched the guardian auto-review preferred model to codex-auto-review for PR review, risk detection, code quality suggestions, and regression checks; priced the same as gpt-5.5.

Supported Depths:LowMediumHighxhigh
≤272KInput$5.00/M·Cached Input$0.50/M·Output$30.00/M
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.

Image Generation Series

Use gpt-image-2 to generate images inside Codex with per-generation quota deduction. See thesetup guide

gpt-image-2

OpenAI's next-generation image generation model for fast, high-quality design images, illustrations, and visual assets. CodexZH currently charges $0.50 per generation, supports 2K output, and image-to-image is not available yet while we improve the workflow.

Capabilities:Text to image
Fixed$0.50 per generation
Model providerOpenAI
Used by platform for trainingNo
Third-party training rulesDisabled by default for model training
Input/output retentionComplete inputs and outputs are not retained long-term by default

Model capability, pricing, context length, rate limits, content safety policy, service availability, and data processing rules may be affected by the model provider. CodexZH does not guarantee permanent availability or fully accurate outputs for any third-party model.