OpenAI GPT-4-0613: 8K context, 13.1 AA Coding, text-only model at $30/$60 per 1M tokens
OpenAI GPT-4-0613 is a specific checkpoint of the GPT-4 large language model released by OpenAI in June 2023. It operates exclusively on text inputs and produces text outputs. The model supports a…
GPT-4-0613 excels at tasks that require deep reasoning, precise language generation, and adherence to complex instructions. Common use cases include generating and reviewing code in languages like Python, JavaScript, and C++, as well as debugging existing codebases. It can analyze and summarize large text corpora, produce structured reports, and draft detailed technical or legal documents. The model is also effective at breaking down multi-step problems into clear, sequential steps, making it useful for planning and problem-solving. While it primarily works in English, it can handle multiple other languages with varying reliability. For tasks that do not require this level of sophistication, smaller or cheaper models may offer a more cost-effective solution.
GPT-4-0613 is a premium model with higher cost per token. It is not ideal for high-volume, low-complexity tasks such as simple text classification, basic chat interactions, or straightforward translation. For such workloads, models like OpenAI's GPT-3.5 Turbo or other lightweight alternatives may deliver adequate performance at a fraction of the cost. Additionally, if your application does not require the full 8192 token context window or if the maximum output length is rarely needed, a cheaper model with smaller context might be more appropriate. Always evaluate the trade-off between output quality and cost for your specific use case to determine whether GPT-4-0613 is justified.
GPT-4-0613 is primarily trained on English but has demonstrated capability in many other languages, including French, Spanish, German, Chinese, and Arabic. Its performance in non-English languages is generally strong for structured tasks like translation, summarization, and question answering, but it may exhibit lower accuracy and less natural phrasing compared to English. The model's strengths in reasoning and instruction-following extend to multilingual contexts, but users should test specific non-English scenarios to confirm suitability. For applications requiring high precision in a language other than English, consider supplementing with language-specific models or additional validation.
GPT-4-0613 respects system messages and user instructions, making it well-suited for applications that require strict adherence to formatting, tone, or structure. It can generate outputs in JSON, markdown, or other specified formats, and can follow multi-turn instructions with maintained context. The model is particularly reliable for tasks that require step-by-step reasoning or chain-of-thought prompting. However, like all large models, it may occasionally misinterpret ambiguous instructions or produce outputs that deviate from the desired format. It is recommended to provide clear, explicit instructions and, when possible, include examples to guide the model's behavior.
The AA Coding benchmark score of 13.1 for GPT-4-0613 indicates its performance on a coding-specific evaluation. While the exact methodology of this benchmark is not detailed in the provided facts, a higher score generally reflects better accuracy and reliability in generating code, solving programming problems, and understanding code structures. This score positions GPT-4-0613 as a strong choice for coding tasks such as bug fixing, algorithm implementation, and code explanation. It is important to note that benchmark scores are indicative and may not fully capture real-world performance on specific coding tasks. Users should evaluate the model on their own datasets for best results.
Latency for GPT-4-0613 is not specified in the provided facts, but as a high-capacity model, it generally has higher response times compared to smaller models like GPT-3.5 Turbo or GPT-4o mini. Inference speed depends on factors such as the length of the input and output, the complexity of the request, and the underlying infrastructure. Through OrcaRouter, latency may also be influenced by network conditions and the provider's serving capacity. For real-time applications where low latency is critical, consider testing the model under your expected load and possibly using a faster, cheaper model for less demanding parts of your workflow.
GPT-4-0613's primary strengths lie in its logical reasoning, ability to follow complex multi-step instructions, and high accuracy in tasks like code generation and data analysis. It produces well-structured, coherent outputs even for long prompts, making it suitable for drafting comprehensive documents or reports. The model demonstrates strong consistency across multiple turns, which is beneficial for conversational agents that need to maintain context. Its benchmark score of 13.1 on AA Coding underscores its reliability for programming-related tasks. These qualities make it a preferred choice for applications where precision and depth are paramount, even at a higher cost.
GPT-4-0613 has several limitations. It is text-only and cannot process images, audio, or video. Its context window is capped at 8192 tokens, which may be insufficient for very long documents or multi-turn conversations with extensive histories. The model can sometimes produce plausible-sounding but incorrect information (hallucinations), particularly on topics outside its training data. Its pricing is relatively high compared to alternatives, making it less economical for high-volume applications. Additionally, being a snapshot from June 2023, it may not be aware of events or knowledge beyond that date. Users should verify critical outputs and consider using retrieval-augmented generation (RAG) to ground responses in up-to-date information.
GPT-4-0613 is priced at the provider's direct rate with zero markup through OrcaRouter. The cost is $30.00 per 1 million input tokens and $60.00 per 1 million output tokens. Input tokens refer to all text supplied in the user's request, including system messages, conversation history, and the current user query. Output tokens are the text generated by the model in response. Both input and output tokens count toward the total usage and are billed accordingly. There are no additional platform fees beyond the per-token rate. Users are billed based on the actual number of tokens processed per API call.
Because output tokens are priced twice as high as input tokens ($60 vs $30 per 1M tokens), applications that generate long responses can become significantly more expensive than those with short outputs. For example, a request with 10,000 input tokens and a 2,000 token output would cost $0.30 for input and $0.12 for output, totaling $0.42. If the output were 8,000 tokens, the cost would rise to $0.30 + $0.48 = $0.78. To manage costs, consider limiting the max_tokens parameter to the minimum necessary length for your use case. Also, shorter system prompts and trimmed conversation histories reduce input token counts.
No information about token caching or volume discounts is provided in the available facts for GPT-4-0613 through OrcaRouter. This means that each call is billed based on the actual tokens processed without any predetermined caching mechanism that would reduce costs for repeated prompts. Users should contact OrcaRouter directly to inquire about potential caching features or enterprise pricing agreements. It is also possible that OrcaRouter may offer cost optimization options not documented here. For now, operate under the assumption that all tokens are charged at the listed per-token rates with no special discounts.
When using GPT-4-0613, the main trade-off is between output quality and cost. For tasks that demand high accuracy and reasoning, the premium price is often justified. However, for simpler or bulk processing, cheaper alternatives like GPT-3.5 Turbo or GPT-4o mini can reduce costs by up to 90%. Also consider the input-output ratio: if your workflow requires very long inputs (e.g., full documents) but short outputs, the input cost will dominate. Conversely, long outputs will increase the output cost quickly. Evaluate whether the model's specific strengths (like coding accuracy) are necessary for your task, and benchmark a subset of your data before committing to large-scale usage.
To call GPT-4-0613 through OrcaRouter, use the OpenAI-compatible API endpoint at base_url https://api.orcarouter.ai/v1. Set the model parameter to "openai/gpt-4-0613" in your request. You will need an API key from OrcaRouter for authentication. The request format follows OpenAI’s chat completions documentation, supporting parameters such as messages (array of system and user messages), temperature (0 to 2), top_p, max_tokens (up to 8192), n (number of completions), stop (sequences to halt generation), and stream (boolean for streaming). Example using Python's openai library: import openai; openai.api_base = "https://api.orcarouter.ai/v1"; openai.api_key = "your_orcarouter_key"; response = openai.ChatCompletion.create(model="openai/gpt-4-0613", messages=[{"role":"user","content":"Hello"}]).
GPT-4-0613 supports standard OpenAI chat completion parameters. Key parameters include temperature (default 1) which controls randomness—lower values make output more deterministic; top_p (default 1) for nucleus sampling; max_tokens (default set by endpoint, maximum 8192) to limit output length; n (number of completions to generate per request); and stop (strings that stop generation). You can also use presence_penalty and frequency_penalty to encourage or discourage topic repetition. Streaming is supported by setting stream=True, which yields tokens incrementally. All parameters behave identically to OpenAI's own API, enabling seamless migration. For best results, set max_tokens to no more than necessary to control costs, and use a temperature between 0 and 0.5 for factual tasks.
Migrating from OpenAI's direct API to OrcaRouter requires only two changes in your code: update the base_url to https://api.orcarouter.ai/v1 and change the model ID to "openai/gpt-4-0613". Your existing authentication logic should be updated to use an OrcaRouter API key instead of an OpenAI key. All familiar parameters (messages, temperature, max_tokens, etc.) remain unchanged. For example, using the openai Python library, set openai.api_base = "https://api.orcarouter.ai/v1" and openai.api_key = "your_orcarouter_key". Then call openai.ChatCompletion.create(model="openai/gpt-4-0613", ...). OrcaRouter's API is designed to be a drop-in replacement, so no further code modifications are necessary. Test with a small request to confirm connectivity and billing.
Rate limits for GPT-4-0613 via OrcaRouter are not specified in the available facts. Authentication requires an API key provided by OrcaRouter. This key should be included in the request header (e.g., Authorization: Bearer <key>). The exact rate limits may depend on your OrcaRouter plan or account tier. For production usage, contact OrcaRouter for details on maximum requests per minute (RPM) and tokens per minute (TPM). In the absence of specific limits, it is advisable to implement exponential backoff in case of rate-limit errors. Also note that your API key should be kept secure and not exposed in client-side code.
GPT-4-0613 is a later snapshot than GPT-4-0314, released in June 2023 versus March 2023. OpenAI has stated that the June update improves reliability, reduces the likelihood of generating forbidden content, and offers better instruction adherence. Both models have the same context window (8192 tokens) and pricing structure. The benchmark score of 13.1 on AA Coding is specific to GPT-4-0613, while GPT-4-0314 may have slightly different performance numbers. In practice, many users report that GPT-4-0613 is more consistent and less prone to hedging or refusing harmless requests. If you are currently using GPT-4-0314, migrating to GPT-4-0613 is straightforward and likely beneficial.
GPT-4o is OpenAI’s multimodal model that can process text, images, and audio, and is faster and cheaper than GPT-4-0613. GPT-4o's pricing is $5 per 1M input tokens and $15 per 1M output tokens, making it significantly more cost-effective. GPT-4-0613, being text-only, cannot handle non-text inputs. However, GPT-4-0613 may offer slightly more deliberate reasoning for complex text-only tasks, as it is the same underlying model as GPT-4 but without multimodal integration. For applications that require image understanding or need lowest latency, GPT-4o is often the better choice. For pure text tasks where maximum accuracy is paramount and cost is less critical, GPT-4-0613 remains a viable option.
GPT-3.5 Turbo is a significantly cheaper and faster alternative to GPT-4-0613, priced at roughly $3 per 1M input tokens and $6 per 1M output tokens. It is optimized for dialog and simple instructions but lacks the depth of reasoning, coding accuracy, and instruction-following capability of GPT-4-0613. The AA Coding benchmark and complex reasoning tasks show a clear performance gap favoring GPT-4-0613. For high-volume applications with straightforward requirements, GPT-3.5 Turbo is more economical. For critical code generation, multi-step analysis, or nuanced language tasks, GPT-4-0613 provides substantial quality improvements that can justify its higher cost. Consider using GPT-3.5 Turbo for simple or iterative subtasks within a larger workflow to balance cost and quality.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="$ORCAROUTER_API_KEY",
)
response = client.chat.completions.create(
model="openai/gpt-4-0613",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)frequency_penaltylogit_biaslogprobsmax_tokenspresence_penaltyseedstopstreamtemperaturetool_choicetoolstop_logprobstop_p| Input / 1M tokens | $30.00 |
| Output / 1M tokens | $60.00 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
GET /api/public/models/openai/gpt-4-0613Open @misc{orcarouter_gpt_4_0613,
title = {openai/gpt-4-0613 API},
author = {openai},
year = {n.d.},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/openai/gpt-4-0613}
}openai. (n.d.). openai/gpt-4-0613 API. OrcaRouter. https://www.orcarouter.ai/models/openai/gpt-4-0613