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

GPT-5.2 (xhigh) leads the supplied quality measurements, while GPT-5 mini (high) offers a much lower listed cost with unresolved availability evidence. The supplied Artificial Analysis snapshot reports an Intelligence Index of 42.2 for GPT-5.2 and 25.3 for GPT-5 mini, plus Math Index scores of 99 and 90.7 respectively. GPT-5 mini is listed at $0.6875 per 1M blended tokens, compared with $4.8125 for GPT-5.2. Both models show 0.3 seconds of latency in the snapshot, while neither has a reported median output speed. Data provided by https://artificialanalysis.ai/. The central selection problem is therefore not simply quality versus price. It is whether the higher measured scores are worth paying for a model whose current official availability is not confirmed. OpenAI’s current model directory does not list either supplied model name, and the current pricing page does not list either supplied API model ID.

GPT-5.2 (xhigh) is the stronger measured choice for demanding reasoning and mathematical work, but GPT-5 mini (high) is the safer economic choice for large request volumes. The supplied data gives GPT-5.2 an Intelligence Index of 42.2 versus 25.3 for GPT-5 mini. Its Math Index is 99 versus 90.7. Those results support choosing GPT-5.2 when answer quality is the main constraint, especially if errors create substantial review or operational cost. They do not prove that GPT-5.2 is better for coding, because the supplied snapshot contains no GPT-5.2 Coding Index. GPT-5 mini has a Coding Index of 15.6, but there is no comparable score for GPT-5.2, so a coding winner cannot be established. The cost difference is substantial: GPT-5 mini is listed at $0.25 per 1M input tokens and $2 per 1M output tokens, while GPT-5.2 is listed at $1.75 and $14. The blended figures are $0.6875 and $4.8125 per 1M tokens. Both models report 0.3 seconds of latency, so the supplied evidence does not establish a latency advantage. Official documentation also leaves an important gap. OpenAI’s model directory does not currently list gpt-5-2, gpt-5-mini, or xhigh as an independent model entry. OpenAI’s pricing page likewise does not confirm current listed pricing for these IDs. That absence makes deployment availability a decision criterion, not an implementation detail.

GPT-5.2 (xhigh) has the stronger supplied reasoning profile, but the evidence does not establish a coding advantage or a speed advantage. The Intelligence Index gap is 42.2 versus 25.3, and the Math Index gap is 99 versus 90.7. For developers, those measurements suggest that GPT-5.2 may be more suitable for tasks where the model must maintain reasoning quality across difficult constraints, calculations, or multi-step decisions. They do not guarantee better results for a specific repository, language, framework, or agent workflow. The supplied evaluation does not include a GPT-5.2 Coding Index. GPT-5 mini records 15.6 on the Coding Index, but that isolated score cannot support a direct coding comparison. A team selecting a model for code generation should therefore run its own repository tasks before treating GPT-5.2 as the coding winner. The latency result is also neutral. Both models are listed at 0.3 seconds, and neither has a median output-tokens-per-second value. The snapshot cannot answer whether one model streams longer answers faster, reaches a complete answer sooner, or performs better under concurrency. OpenAI’s current model directory provides only broad statements about capabilities for current OpenAI models, including text and image inputs, text outputs, multilingual use, and vision. It does not explicitly attribute those statements to these historical or supplied model names. The practical conclusion is narrow: GPT-5.2 has stronger supplied intelligence and math results, while coding and throughput remain unproven.

GPT-5 mini (high) is the clear listed-price choice, but GPT-5.2 (xhigh) can still be economically rational when higher answer quality reduces downstream work. GPT-5 mini costs $0.25 per 1M input tokens and $2 per 1M output tokens in the supplied snapshot. GPT-5.2 costs $1.75 and $14. The blended figures are $0.6875 for GPT-5 mini and $4.8125 for GPT-5.2. The price gap matters most for workloads with high request volume, long prompts, or verbose outputs. It matters less when a single wrong answer triggers manual review, a failed build, a retry, or an incident. That is an economic hypothesis, not a measured result in the supplied materials. The data does not provide task success rates, retry rates, token distributions, or total workflow costs, so no break-even point can be calculated responsibly. The cheaper model can also become more expensive if it requires more validation passes, but the supplied evidence does not prove that GPT-5 mini requires more passes. The reverse is possible too: GPT-5.2 may add cost without improving the particular task. OpenAI’s current pricing page does not list either supplied model ID, so the figures should be treated as comparison data from the provided snapshot rather than confirmed current OpenAI prices. Before production commitment, verify the actual endpoint price and availability, then measure cost per accepted result rather than cost per generated token.

GPT-5 mini (high) is the default choice for cost-sensitive scale, while GPT-5.2 (xhigh) is the targeted choice for quality-sensitive reasoning tasks. Choose GPT-5.2 when the supplied Intelligence Index of 42.2 and Math Index of 99 align with the work being evaluated, and when the additional listed price is acceptable. Suitable candidates include complex analysis, mathematical reasoning, and tasks where review effort is expensive. The supplied evidence does not confirm that GPT-5.2 is better for coding, so do not select it for software engineering solely from its overall score. Choose GPT-5 mini when request volume and predictable token economics dominate. Its supplied blended price is $0.6875 per 1M tokens, with $2 per 1M output tokens. That makes it the better starting point for broad automation, lightweight transformations, and workloads where occasional quality differences are cheap to detect and correct. This recommendation remains conditional because OpenAI’s model directory does not currently confirm either supplied model as a listed model entry. The safest deployment path is to verify the exact API ID, run a representative task set, and compare accepted-output cost. The most important missing experiment is a paired coding evaluation, because GPT-5 mini has a Coding Index of 15.6 while GPT-5.2 has no supplied coding score. Teams should also test long-context behavior, tool calls, structured output, retries, and concurrency because the supplied briefs provide no confirmed context window, output limit, API parameter mapping, or throughput result for either model. Those gaps prevent a fully confident production recommendation.

GPT-5.2 (xhigh) has the larger evidence gap because its supplied identity, availability, and coding performance remain unconfirmed in current official pages. GPT-5 mini also lacks a current official catalog entry, so the same deployment caution applies to both models. OpenAI’s model directory does not list GPT-5.2, gpt-5-2, gpt-5-mini, or xhigh in the supplied research. The research also found no official release announcement or model-specific capability page for GPT-5.2. For GPT-5 mini, the research found no official explanation connecting the display name “GPT-5 mini (high)” to a stable API model ID or confirming that high is an API parameter value. No reliable Reddit, Hacker News, or X posts were found for either supplied model, so community claims about coding feel, quirks, or speed cannot be treated as evidence. The briefs also provide no verified context window, output limit, tool support, failure pattern, or deprecation status for either model. Developers should resolve these questions before building an integration around a stable identifier. The supplied release dates are 2025-12-11 for GPT-5.2 and 2025-08-07 for GPT-5 mini, but the current official pages do not establish the models’ present service status. The result is a comparison of supplied benchmark and price data, not proof of current product availability.