OpenAI GPT-5.1-2025-11-13: 400k context, AA Math 94.0, via OrcaRouter API
OpenAI GPT-5.1-2025-11-13 is a capable model from OpenAI's GPT-5 series, timestampted November 13, 2025. It supports a context window of 400,000 tokens and a maximum output of 128,000 tokens. The…
The model excels in mathematical reasoning, as evidenced by its AA Math score of 94.0. It can perform complex calculations, solve multi-step word problems, and work with advanced mathematical concepts. Beyond math, it supports general natural language understanding and generation across domains. The multimodal capability allows it to interpret images and files, making it useful for tasks that combine visual and textual information, such as explaining a graph or extracting data from a scanned document. It also supports code generation, translation, summarization, and creative writing, consistent with other GPT-5-level models.
While GPT-5.1-2025-11-13 offers high capabilities, it is priced higher than many alternative models: $1.25 per 1M input tokens and $10.00 per 1M output tokens. For tasks that do not require the full 400k context window, advanced math reasoning, or multimodal input, a smaller and cheaper model available on OrcaRouter might be more cost-effective. For example, if your task is simple text classification, short-form chat, or standard summarization, consider models like GPT-4o-mini or Claude 3 Haiku to reduce costs. Additionally, if latency is a concern, smaller models typically respond faster. Use this model for high-stakes reasoning tasks or when you need the large context window.
Yes, the model accepts file input modality in addition to image and text. Files can be uploaded directly as part of the API request, and the model will read and process their content. Supported file types typically include PDF, Word documents, plain text, and possibly spreadsheets, though the exact list depends on OpenAI's implementation. The model can extract text from these files and integrate that information into its reasoning. This is useful for tasks like summarizing a PDF report, answering questions based on a research paper, or analyzing tabular data. However, the model might not handle highly formatted or scanned documents perfectly; consider preprocessing for best results.
The AA Math (Advanced Automated Math) evaluation tests a model's ability to solve complex mathematical problems across various subfields, including algebra, calculus, statistics, and geometry. A score of 94.0 indicates that the model correctly answered 94% of the benchmark questions, placing it among the top-performing models for mathematical reasoning. In practice, this means the model can reliably handle challenging math tasks such as multi-step proofs, applied problem solving, and physics-related calculations. However, no model is perfect, and users should verify critical results, especially for novel or open-ended mathematical problems. The score is based on a specific test set and may not generalize to all real-world tasks.
Exact latency figures are not provided in the available data, but generally, models with large context windows and high output limits have longer inference times due to increased memory and computation requirements. The actual speed depends on factors like input and output token count, server load, and API infrastructure. OrcaRouter's API provides standard OpenAI-compatible endpoints; you can expect latency in the range of several seconds to tens of seconds for long generations. Streaming responses can reduce time-to-first-token. For latency-sensitive applications, consider testing with small inputs first, or use a faster smaller model if speed is critical.
Despite its high math benchmark score, the model may still produce errors on rare or extremely complex problems, especially those requiring precise intermediate steps. The large context window does not guarantee perfect recall of information from the very beginning of the input; models can exhibit recency bias. Multimodal understanding, while present, may struggle with ambiguous or low-quality images. The pricing is higher than many alternatives, so it may not be cost-effective for simple tasks. Additionally, as a snapshot model from November 2025, it may not incorporate knowledge or events after that date. Using the model through OrcaRouter does not change these underlying limitations.
Pricing for this model is $1.25 per 1 million input tokens and $10.00 per 1 million output tokens. These rates are billed at the provider rate with zero markup, meaning you pay exactly what OpenAI charges—OrcaRouter adds no additional fee. This is beneficial for users who want predictable, transparent pricing. Note that token counts include both prompt and generation. If you use a large context window of 400k tokens as input, the cost for that single request would be $0.50 (400k tokens at $1.25/M). Output costs add separately. For comparison, smaller models on OrcaRouter may cost a fraction of this per token.
The primary trade-off is the higher per-token cost relative to smaller models. While the model offers advanced capabilities, users should estimate typical input and output token volumes to decide if the expense is justified. For example, if you frequently generate long outputs (e.g., 50k tokens), the output cost at $10/M would be $0.50 per request. For high-volume applications, costs can accumulate quickly. Consider caching responses where possible to avoid repeated processing. OrcaRouter does not offer additional discounts or special tiers for this model; pricing is straightforward. If budget is tight, explore cheaper alternatives like GPT-4o-mini (if available) or other providers on OrcaRouter.
OrcaRouter provides standard API access but does not inherently cache responses; caching is the responsibility of the user. You can implement your own caching layer for identical requests to reduce token usage and cost. OrcaRouter’s API is stateless—each request is processed independently. For long-running projects, consider using prompt caching techniques such as splitting large contexts wisely or reusing embeddings. There are no special billing features like batch discounts for this model. The zero-markup pricing simplifies budgeting but does not include built-in optimization tools.
You access the model through OrcaRouter's OpenAI-compatible API. Set the base URL to https://api.orcarouter.ai/v1. Use the model ID "openai/gpt-5.1-2025-11-13" in your request body. The API supports standard OpenAI parameters such as messages, max_tokens, temperature, top_p, etc. For example, to create a chat completion, make a POST request to /chat/completions with the model parameter set to "openai/gpt-5.1-2025-11-13". You can also include image or file content in the messages using the appropriate content types (e.g., image_url). Ensure you have a valid API key from OrcaRouter for authentication.
Common parameters include max_tokens (up to 128,000), temperature (0-2 for randomness), top_p (nucleus sampling), frequency_penalty, presence_penalty, stop sequences, and stream (boolean for streaming). The context window is 400,000 tokens, so ensure your input does not exceed that total (including system and assistant messages). You can also specify the response_format parameter if supported (e.g., json_object) to force structured output. OrcaRouter passes these parameters directly to the underlying OpenAI model. For multimodal content, use the 'content' array in messages with parts containing text and image/file data.
Yes, migration is straightforward because OrcaRouter offers an OpenAI-compatible API. Most existing code that uses the OpenAI Python or Node SDK can switch by changing the base_url and API key. The model ID on OrcaRouter is "openai/gpt-5.1-2025-11-13" (note the prefix). No other changes to request bodies or response handling are needed. If you were using a different provider's API that also follows OpenAI standards, you can simply update the endpoint. OrcaRouter provides zero-markup pricing, so costs may be similar or lower depending on previous markups.
Compared to earlier GPT-4 models, GPT-5.1 offers a larger context window (400k vs 128k typically), higher output limit (128k vs 4k-32k), and significantly better AA Math score (94.0 vs lower). It also supports image and file inputs, which GPT-4 lacked natively. However, it is more expensive per token than GPT-4o or GPT-4o-mini. Compared to GPT-5.0 (if available), this snapshot may have improvements but details are not public. For users who need fewer tokens or lower cost, GPT-4o-mini provides a faster, cheaper alternative with reasonable quality.
Comparisons depend on specific benchmarks. The AA Math score of 94.0 is a strong indicator for reasoning, but other models like Claude 3.5 Sonnet excel in coding and nuanced safety. The context window of 400k matches or exceeds most competitor models (e.g., Claude 3.5 Sonnet offers 200k). Modal capabilities vary; some competitors also handle images and files. Pricing: GPT-5.1 at $1.25/1M input is more expensive than some, but in line with premium offerings. For mathematical tasks, this model may outperform; for creative writing or summarization, alternatives might suffice. Users should evaluate on their specific use case.
It can be used in production given the high context and strong reasoning, but cost is a key factor. At $10/1M output tokens, high-volume generation can become expensive. The model is accessed through OrcaRouter's reliable API, which provides standard uptime and performance. However, for extreme scalability, consider implementing caching, prompt optimization, and possibly using a cheaper model for simpler sub-tasks. The model's latency may also be higher than smaller alternatives. If your application demands real-time responses, test thoroughly. Overall, it's a premium choice for premium tasks.
Key differences include: larger context window (400k vs typically 128k), longer max output (128k vs 16k for GPT-4o), higher AA Math score (94.0 vs lower), and support for file input modality besides image and text. Pricing is higher for GPT-5.1: $1.25/$10 per M tokens vs approximately $2.50/$10 for GPT-4o (exact depends on model). So GPT-5.1 is cheaper in input but same in output? Actually GPT-4o input is $2.50/M, output $10/M, so GPT-5.1 input is cheaper. But GPT-4o has faster speeds typically. For multimodal and math-intensive tasks requiring large context, GPT-5.1 is likely superior.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="$ORCAROUTER_API_KEY",
)
response = client.chat.completions.create(
model="openai/gpt-5.1-2025-11-13",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)include_reasoningmax_completion_tokensmax_tokensreasoningresponse_formatseedstreamstructured_outputstool_choicetools| Input / 1M tokens | $1.25 |
| Output / 1M tokens | $10.00 |
| Cache read / 1M | $0.125 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
GET /api/public/models/openai/gpt-5.1-2025-11-13Open @misc{orcarouter_gpt_5_1_2025_11_13,
title = {openai/gpt-5.1-2025-11-13 API},
author = {openai},
year = {n.d.},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/openai/gpt-5.1-2025-11-13}
}openai. (n.d.). openai/gpt-5.1-2025-11-13 API. OrcaRouter. https://www.orcarouter.ai/models/openai/gpt-5.1-2025-11-13