Loading page
OpenAI
GPT-5.1 Codex mini
GPT-5.1 Codex mini is a smaller, coding-oriented variant of OpenAI's GPT-5.1 Codex line, positioned for software engineering tasks where a lighter model is preferred.
Input from
$0.250 / 1M tokens
across 1 provider
API Pricing
| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.250 | $2.00 | $0.030 |
Prices updated daily. Last check: Sep 5, 2026
GPT-5.1 Codex mini pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
24.0 / 100
Math
91.7 / 100
Reasoning & Knowledge
- MMLU-Pro82.0%
- GPQA Diamond81.3%
- Humanity's Last Exam18.5%
Coding
- LiveCodeBench83.6%
Math
- AIME 202591.7%
Agentic & Tool Use
- Terminal-Bench Hard33.3%
- τ²-bench62.9%
Instruction & Long Context
- IFBench67.9%
- Long-Context Reasoning66.7%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- OpenAI
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Coding-specialized variant: the Codex branch is tuned for software engineering tasks rather than general chat
- Mini tier within GPT-5.1 Codex, aimed at high-volume or latency-sensitive coding turns
- Same GPT-5.1 generation as its larger Codex sibling, so escalation between them is a model-string change
- Chat-style API surface that drops into existing OpenAI-compatible agent loops and IDE integrations
- Suited to repetitive agent steps (edits, test runs, review passes) where a smaller model reduces per-turn overhead
- Available from multiple inference providers — compare current options in the pricing table on this page
Limitations
- We do not have a confirmed context window on file for this variant, which matters for whole-repository tasks
- No populated throughput or time-to-first-token measurements in our benchmark record
- Mini-tier models generally trade some depth on complex multi-file reasoning compared with the full Codex model
- Coding-tuned models are often a weaker fit for open-ended writing, summarization, or general conversational work
- Modality support beyond text is not something we track for this model
Key Features
•Coding-oriented tuning within the GPT-5.1 Codex line
•Mini tier positioned below the full GPT-5.1 Codex model
•Chat-completion API interface
•Designed for use inside coding agents, CLI tools, and IDE integrations
•Part of OpenAI's GPT-5.1 generation
•Availability across multiple inference providers tracked on this page
About GPT-5.1 Codex mini
GPT-5.1 Codex mini is a chat model from OpenAI that belongs to the GPT-5.1 Codex branch of the GPT-5.1 family. The Codex naming indicates a variant tuned for software engineering work — reading and editing code, following repository conventions, and operating inside coding agents and CLI tools — rather than a general-purpose assistant profile. The "mini" designation places it below the full Codex model in the same generation, in the slot typically reserved for higher-volume or latency-sensitive workloads.
Because it is exposed as a chat-completion style model, it is used through the same request/response patterns as other OpenAI models, which makes it straightforward to slot into an existing agent loop or IDE integration that already targets the GPT-5.1 line. We do not currently track a confirmed context window, modality list, or independent throughput and latency measurements for this specific variant; the benchmark record on file has no populated speed figures, so any performance expectations should be validated against your own workload rather than inferred from a sibling model.
In practice, mini-tier coding models are chosen when a team is running many short agent turns — file edits, test fixes, lint and refactor passes, code review comments — and wants to reserve the larger Codex model for the harder planning and multi-file reasoning steps. Comparing GPT-5.1 Codex mini against the full GPT-5.1 Codex, and against non-Codex GPT-5.1 variants, on your own repository is the most reliable way to decide where it fits.
Common Use Cases
GPT-5.1 Codex mini fits the repetitive inner loop of software development work: applying targeted file edits, fixing failing tests, resolving lint and type errors, drafting small functions, writing unit tests, and generating code review comments. Its mini tier makes it a candidate for agent architectures that fire many short turns per task, where a larger Codex model is reserved for planning, architecture decisions, or cross-file refactors that need more reasoning depth. It is also a reasonable default for developer-tooling features that run continuously — inline completions triggered by editor events, CI-side patch suggestions, or batch codemod runs across many files. For work that is primarily prose, analysis, or open-ended conversation, a general-purpose GPT-5.1 variant is usually the better match.
Frequently Asked Questions
How much does GPT-5.1 Codex mini cost?
Pricing depends on the provider you use and the pricing model — per-input-token, per-output-token, or hosted endpoint arrangements all differ, and rates change frequently. Check the pricing table on this page for the current tracked rates rather than relying on a figure quoted in an article.
What is GPT-5.1 Codex mini best used for?
Software engineering tasks that run at volume: code edits, test generation and repair, refactor passes, lint fixes, and code review comments inside an agent or IDE integration. It is a coding-branch model, so it is a better match for those workloads than for general writing or chat.
How does it differ from the full GPT-5.1 Codex model?
Both are in the GPT-5.1 Codex branch from OpenAI, but the mini tier sits below the full model. Mini-tier variants are generally chosen for shorter, more frequent turns, while the larger Codex model is used for the harder planning and multi-file reasoning steps. Many teams route between the two within the same agent.
What context window does GPT-5.1 Codex mini support?
We do not have a confirmed context window recorded for this variant, so we do not publish a number. Check OpenAI's model documentation or your provider's model card before designing a workflow that depends on very large inputs.
Are there published speed benchmarks for this model?
Our benchmark record for GPT-5.1 Codex mini sourced from Artificial Analysis has no populated output-tokens-per-second or time-to-first-token values. Measure latency and throughput on your own prompts and provider before committing to it in a latency-sensitive path.