OpenAI's gpt-3.5-turbo-1106 via OrcaRouter: 16K context, 46.2 MMLU-Pro, $1/$2 per 1M tokens, zero-markup pricing.
gpt-3.5-turbo-1106 is a specific snapshot of OpenAI's GPT-3.5 Turbo model, made available in November 2023. It is a text-only model, meaning it accepts and generates only text inputs and outputs, not…
This model excels at general-purpose text generation tasks: answering questions, engaging in dialogue, summarizing documents, translating text, generating creative writing, and providing explanations. It supports function calling, allowing it to output structured data and interact with external tools or APIs based on provided function schemas. It can also handle multi-turn conversations with a context window of up to 16K tokens. For tasks that require high accuracy on factual queries, code generation, or structured reasoning, it performs well but may occasionally produce less precise outputs than larger models.
Function calling in gpt-3.5-turbo-1106 allows the model to generate structured JSON outputs based on function definitions you supply in the request. When you include a list of functions with names, descriptions, and parameter schemas, the model can decide to call one or more functions by returning a function_call object. This is useful for building agents that need to query databases, call APIs, or perform actions. The feature was improved in the 1106 snapshot compared to earlier versions, making it more reliable and reducing redundant tool usage. OrcaRouter supports all OpenAI-compatible function calling parameters.
If your use case involves very simple tasks with very high volume, you might consider an even cheaper model like gpt-3.5-turbo-0125 (same price but newer) or smaller open-source models offered through OrcaRouter. On the other hand, if you need significantly better reasoning, factuality, multimodal input (images, audio), or a larger context window (128K tokens), consider upgrading to gpt-4o or gpt-4-turbo. gpt-3.5-turbo-1106 is a balanced choice for most text-only applications where cost-per-token is a primary concern.
MMLU-Pro is a variant of the Massive Multitask Language Understanding benchmark that includes 57 subjects with a more challenging set of questions. A score of 46.2 indicates that the model correctly answered 46.2% of the test questions. This is a moderate score, reflecting the model's general knowledge across diverse domains. For comparison, larger models like GPT-4 typically score above 80 on similar benchmarks. This score helps set expectations: gpt-3.5-turbo-1106 is capable but not state-of-the-art for deep reasoning or specialized knowledge.
No other benchmark scores are provided for this model, but publicly available information shows that gpt-3.5-turbo-1106 performs strongly on MMLU, HumanEval (code generation), and summarization tasks relative to its size class. It is generally considered to be among the best performing open-weight models for its cost tier. Users should expect good performance on common NLP tasks but acknowledge that it may lag behind GPT-4 on complex reasoning, math, and multi-step instructions. For factual consistency, the model may produce plausible but incorrect answers, so verification is recommended for critical applications.
Specific latency numbers are not provided. However, as a text-only model with a relatively small parameter count (compared to GPT-4), gpt-3.5-turbo-1106 generally offers low latency output, often completing short responses in under a second under moderate load. Throughput can be high, making it suitable for real-time applications when combined with streaming. OrcaRouter's infrastructure ensures reliable connection to OpenAI's endpoints. For the most accurate latency expectations, users should benchmark their own use cases using the OrcaRouter API.
Key limitations include: a 4,096 token maximum output, which restricts generation length; no multimodal capabilities (cannot process images or audio); a tendency to hallucinate on obscure facts; and limited reasoning depth for complex problems. The model's 16K context window, while generous, is smaller than some alternatives (e.g., 128K). It also has a lower knowledge cutoff than later snapshots (January 2023). On adversarial or ambiguous prompts, it may produce biased or unsafe outputs. Developers should implement appropriate guardrails and human oversight for production use.
Pricing for openai/gpt-3.5-turbo-1106 through OrcaRouter is billed at the provider rate with zero markup. Specifically, input tokens cost $1.00 per 1 million tokens, and output tokens cost $2.00 per 1 million tokens. This applies to all requests, including streaming. There are no additional platform fees. Charges are computed based on the number of tokens in the prompt (input) and the generated response (output). Both cached and uncached requests are billed at the same rate. You can track usage via the OrcaRouter dashboard.
For applications that require many short interactions (e.g., chatbot conversations), the cost per message is low because each request consumes few tokens. For tasks that involve long prompts or generate long responses, the cost scales linearly with token count. Compared to GPT-4 (which can be 10-30x more expensive), gpt-3.5-turbo-1106 is economical for volume. However, if a task could be solved with fewer tokens using a cheaper model, that may be even more cost-effective. OrcaRouter's zero-markup pricing ensures you only pay what OpenAI charges.
No specific caching or discount programs are mentioned for this model on OrcaRouter. All requests are billed at the standard provider rate. Some platforms offer prompt caching to reduce costs for repeated inputs, but OrcaRouter's documentation does not indicate such a feature for this model. Users should plan their token usage accordingly. For very high volume, it may be worth contacting OrcaRouter directly to inquire about volume discounts, though none are advertised. The zero-markup pricing itself already provides a transparent cost baseline.
To make a request, send a POST request to https://api.orcarouter.ai/v1/chat/completions with a JSON body containing the model field set to 'openai/gpt-3.5-turbo-1106', and a messages array (with role and content). Include your OrcaRouter API key in the Authorization header (Bearer <key>). Optional parameters include temperature, max_tokens, top_p, frequency_penalty, presence_penalty, stop, stream, and functions. The response follows OpenAI's chat completion format. Example using Python openai library: client = OpenAI(base_url='https://api.orcarouter.ai/v1', api_key='your_key').
This model supports the same parameters as OpenAI's gpt-3.5-turbo-1106. Key parameters include: messages (required array of objects with role, content, optionally name), max_tokens (default infinite but capped at 4096), temperature (0-2, default 1), top_p (0-1, default 1), n (number of choices, default 1), stream (boolean), stop (string or array up to 4 strings), presence_penalty (-2 to 2), frequency_penalty (-2 to 2), logit_bias (map of token IDs to bias), user (string for end-user identification), functions (array of function objects), function_call (auto/none or specific function). All are supported through OrcaRouter.
Migration requires only two changes to your existing OpenAI-based code. First, change the base URL from https://api.openai.com/v1 to https://api.orcarouter.ai/v1. Second, replace your OpenAI API key with an OrcaRouter API key. The request and response formats are identical, so no changes to the payload structure are needed. After updating these two settings, all existing logic for chat completions, function calling, and streaming will continue to work. You can also use the same OpenAI client libraries (e.g., Python openai, Node.js openai) by passing the new base_url and api_key.
OrcaRouter returns standard HTTP status codes and error messages compatible with OpenAI's API. Common errors: 401 (unauthorized), 429 (rate limit exceeded), 500 (server error). Rate limits are enforced by OrcaRouter based on your plan; for details, refer to OrcaRouter documentation. Since the underlying model is hosted by OpenAI, errors from the provider are relayed transparently (e.g., 400 for invalid parameters, 503 for overload). It is recommended to implement exponential backoff for 429 and 5xx errors. Streaming errors are signaled by a truncated stream; listen for finish_reason in the final chunk.
The 0125 snapshot is a newer version of GPT-3.5 Turbo, released in January 2025. It offers the same pricing and text-only modality but includes a more recent knowledge cutoff and possibly improved consistency and factual accuracy. Both models share the same 16K context window and 4K output limit. gpt-3.5-turbo-0125 is generally recommended over the 1106 version for new projects due to its freshness, but the 1106 snapshot remains available for compatibility reasons. No benchmark scores are provided for either model beyond 46.2 MMLU-Pro for 1106; users may test both for their specific use case.
gpt-4o is OpenAI's more capable model that supports text, image, and audio inputs, and has a 128K context window and 16K max output tokens. It scores significantly higher on reasoning benchmarks and is more reliable for complex tasks. However, it is substantially more expensive: roughly 10-15x the cost per token of gpt-3.5-turbo-1106. For applications where high accuracy and multimodality are not required, gpt-3.5-turbo-1106 offers better cost efficiency. If you need the advanced capabilities, upgrading to gpt-4o through OrcaRouter is straightforward by changing the model ID to 'openai/gpt-4o'.
Comparable models from other providers include Anthropic's Claude 3 Haiku (text-only, similar speed and cost) and Google's Gemini 1.5 Flash. Each has different API formats but can be accessed via OrcaRouter. gpt-3.5-turbo-1106 is widely adopted and integrated into many tools. Its function calling support is mature. However, models like Claude 3 Haiku may offer larger context windows (200K) at similar pricing, while Gemini 1.5 Flash provides multimodal capabilities. The choice depends on ecosystem compatibility, specific benchmark performance, and latency requirements. OrcaRouter allows you to try different models under the same API format.
Choose gpt-3.5-turbo-1106 when you need a reliable, cost-effective text model that is widely compatible with existing tools and supports function calling. It is ideal for high-volume applications where each request is relatively short and does not require deep reasoning. Avoid it if you need multimodal inputs, a larger context window, or higher factual accuracy on complex topics. If you need the latest knowledge cutoff, prefer gpt-3.5-turbo-0125. For ultimate performance, select gpt-4o. OrcaRouter's model catalog allows easy switching between these options without code changes.
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-3.5-turbo-1106",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)frequency_penaltylogit_biaslogprobsmax_tokenspresence_penaltyresponse_formatseedstopstreamtemperaturetool_choicetoolstop_logprobstop_p| Input / 1M tokens | $1.00 |
| Output / 1M tokens | $2.00 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
GET /api/public/models/openai/gpt-3.5-turbo-1106Open @misc{orcarouter_gpt_3_5_turbo_1106,
title = {openai/gpt-3.5-turbo-1106 API},
author = {openai},
year = {n.d.},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/openai/gpt-3.5-turbo-1106}
}openai. (n.d.). openai/gpt-3.5-turbo-1106 API. OrcaRouter. https://www.orcarouter.ai/models/openai/gpt-3.5-turbo-1106