GPT-5.2 (medium) vs GPT-5 mini (high): The Ultimate Performance & Pricing Comparison

GPT-5.2 (medium) leads the supplied intelligence and math evaluations, while GPT-5 mini (high) costs far less to operate. The data snapshot reports scores of 38 and 96.7 for GPT-5.2 (medium), compared with 25.3 and 90.7 for GPT-5 mini (high) on those two indexes. It also reports blended token prices of $4.8125 and $0.6875 respectively. Both models have a reported latency of 0.3 seconds, and neither has a reported median output speed in the supplied data.\n\nThe central selection problem is not simply quality versus price. The current OpenAI model directory does not list either exact model identity, and the OpenAI pricing page does not list either model’s current price. That creates a verification risk for any production decision.\n\nData provided by https://artificialanalysis.ai/

GPT-5.2 (medium) is the stronger measured model, but GPT-5 mini (high) is the more economical choice when the workload tolerates weaker evaluation results. The supplied Artificial Analysis data gives GPT-5.2 (medium) an Intelligence Index of 38 versus 25.3 for GPT-5 mini (high). GPT-5.2 (medium) also leads the Math Index at 96.7 versus 90.7. Those results support a quality advantage for reasoning-heavy tasks, but they do not establish a universal winner for coding because the supplied coding evaluation contains a result only for GPT-5 mini (high), at 15.6.\n\nThe commercial gap is much wider than the reported evaluation gap. GPT-5 mini (high) is listed at $0.25 per 1M input tokens and $2 per 1M output tokens, while GPT-5.2 (medium) is listed at $1.75 and $14. The blended figures are $0.6875 and $4.8125. This makes GPT-5.2 (medium) difficult to justify for high-volume, low-risk requests unless its quality advantage reduces retries, review time, or downstream failures.\n\nAvailability remains unresolved. The OpenAI model directory does not provide an exact entry for either requested model name. The OpenAI pricing page also does not confirm the supplied prices as current public API prices.\n\nThe evidence therefore supports a conditional decision: choose GPT-5.2 (medium) for higher-stakes reasoning if access is verified, and choose GPT-5 mini (high) for cost-sensitive workloads if its API identity and behavior are verified.

GPT-5.2 (medium) is the measured quality leader for general intelligence and mathematics, but the available evidence cannot prove a coding advantage. The Intelligence Index is 38 for GPT-5.2 (medium) and 25.3 for GPT-5 mini (high), while the Math Index is 96.7 and 90.7. These results suggest that GPT-5.2 (medium) deserves priority for tasks where planning, multi-step reasoning, or mathematical reliability dominate the outcome. The supplied data does not identify which individual task types drive those scores.\n\nGPT-5 mini (high) has a Coding Index value of 15.6, but no corresponding GPT-5.2 (medium) value appears in the snapshot. A responsible comparison cannot convert that missing value into a win for either model. Developers should treat coding quality as an open question and test repository-specific tasks, especially code modification, debugging, test generation, and tool-using workflows. No reliable community posts with disclosed methods were found for either exact model identity.\n\nLatency does not separate the models in the supplied comparison. Each reports 0.3 seconds, and neither reports a median output rate. That means the data cannot show whether one model streams longer answers more efficiently or provides better time to first useful token.\n\nThe OpenAI model directory gives only broad descriptions of current model capabilities. It does not verify that those capabilities apply to either requested model identity, nor does it provide model-specific limits, API parameters, or benchmark results.

GPT-5 mini (high) is the economic default because its reported token prices are materially lower, but GPT-5.2 (medium) can still be cheaper at the application level if it prevents expensive retries and human review. The snapshot lists GPT-5 mini (high) at $0.25 per 1M input tokens and $2 per 1M output tokens. GPT-5.2 (medium) is listed at $1.75 and $14. The blended prices are $0.6875 and $4.8125.\n\nThe practical effect depends on response shape. Output-heavy workflows are especially exposed to the gap because the reported output prices are $2 for GPT-5 mini (high) and $14 for GPT-5.2 (medium). A coding agent that repeatedly emits long patches, explanations, or failed attempts may make the premium visible quickly. A short classification or routing request may not create enough quality value to justify the higher-priced model.\n\nThe cheaper option becomes more expensive when weaker results trigger additional calls, manual correction, test failures, or escalation to a stronger model. The supplied data does not include token distributions, retry rates, task success rates, or review costs, so it cannot determine the true cost per successful task.\n\nThe OpenAI pricing page does not currently list either exact requested model. It also does not verify whether the snapshot prices represent current standard, Batch, Flex, or Fast mode prices. Cost planning should therefore treat the supplied values as comparison data, not as a confirmed purchasing commitment.

GPT-5 mini (high) is the better first choice for high-volume developer features, while GPT-5.2 (medium) is the better candidate for difficult reasoning if production access can be confirmed. Use GPT-5 mini (high) for request classification, lightweight code explanation, routine transformations, and workflows where a lower token bill matters more than peak evaluation scores. Its reported blended price is $0.6875, and its latency is 0.3 seconds.\n\nUse GPT-5.2 (medium) for complex planning, mathematical reasoning, architectural trade-off analysis, and tasks where an incorrect answer creates substantial downstream work. Its reported Intelligence Index of 38 and Math Index of 96.7 are the strongest supplied quality signals in this comparison. The premium is harder to defend for routine traffic because its blended price is $4.8125.\n\nDo not choose either model solely because the interface labels one variant “high” or “medium.” The OpenAI model directory does not confirm that GPT-5.2 (medium) maps to gpt-5-2-medium, or that GPT-5 mini (high) maps to gpt-5-mini. The directory also does not confirm whether “high” is a model name or an API reasoning parameter.\n\nA staged selection is the safest conclusion. Verify the exact model ID, availability, and effective price first. Then evaluate representative coding tasks because the supplied coding evidence is incomplete. Select GPT-5 mini (high) when its task success is acceptable. Escalate only the difficult cases to GPT-5.2 (medium) if the measured quality gain offsets its higher price.

GPT-5.2 (medium) should not enter production until its exact API identity and availability are verified through current OpenAI documentation. The OpenAI model directory contains no dedicated entry for GPT-5.2 (medium) or gpt-5-2-medium, and the OpenAI pricing page contains no current listing for that identifier.\n\nGPT-5 mini (high) requires the same verification because the requested display name may combine a model identity with a setting rather than describe a standalone model. The supplied research found no official confirmation that “high” is an API model variant or a supported reasoning parameter for gpt-5-mini.\n\nThe benchmark evidence is useful for prioritizing tests, not for closing every selection question. GPT-5.2 (medium) has stronger supplied intelligence and math results, but the coding comparison is incomplete. Neither model has a supplied context window, output limit, median output speed, or verified model-specific failure profile.\n\nThe most important missing evidence is production-oriented: exact endpoint behavior, tool support, context limits, coding success on the target repositories, and cost per successful task. Those gaps prevent a definitive recommendation independent of workload and account access.