{"id":15546,"date":"2026-11-04T21:14:51","date_gmt":"2026-11-04T13:14:51","guid":{"rendered":"https:\/\/imastudio.com\/?p=15546"},"modified":"2026-11-04T21:14:51","modified_gmt":"2026-11-04T13:14:51","slug":"gpt-6-1-terra","status":"publish","type":"post","link":"https:\/\/imastudio.com\/ko\/blog\/gpt-6-1-terra","title":{"rendered":"Running Out of AI Quota? What GPT-6.1 Terra&#8217;s Published Numbers Say"},"content":{"rendered":"<style>\n#tj{--bg:#faf9f6;--ink:#1c1d2b;--mute:#6b6a78;--line:#e3e0ea;--pu:#6f5fa8;--pu-d:#54478a;--lil:#ece9f3;--card:#fff;--gold:#b8924a;--coral:#c8573e;--green:#2f7d5b;--serif:Georgia,\"Noto Serif SC\",\"Songti SC\",\"STSong\",\"Times New Roman\",serif;--sans:-apple-system,BlinkMacSystemFont,\"Segoe UI\",\"PingFang SC\",\"Helvetica Neue\",Arial,sans-serif}\n#tj, #tj *{box-sizing:border-box;margin:0;padding:0}\n#tj{background:transparent;color:var(--ink);font:17px\/1.85 var(--sans);-webkit-font-smoothing:antialiased}\n#tj a{color:inherit}\n#tj .btn{display:inline-block;background:var(--ink);color:#fff;text-decoration:none;font-weight:600;font-size:14px;padding:11px 20px;border-radius:6px;transition:.2s}\n#tj .btn:hover{background:var(--pu-d)}\n#tj .btn.line{background:transparent;color:var(--ink);border:1px solid var(--ink)}\n#tj .btn.line:hover{background:var(--ink);color:#fff}\n#tj .hero{padding:72px 0 56px;border-bottom:1px solid var(--line)}\n#tj .kicker{font-size:12px;letter-spacing:.2em;text-transform:uppercase;color:var(--pu);font-weight:700;margin-bottom:22px}\n#tj .kicker i{display:inline-block;width:7px;height:7px;border-radius:50%;background:var(--pu);margin-right:10px}\n#tj .hero h1{font-family:var(--serif);font-size:clamp(38px,6.2vw,76px);line-height:1.06;letter-spacing:-2px;font-weight:700;max-width:960px}\n#tj .hero h1 em{font-style:italic;color:var(--pu)}\n#tj .hero .dek{font-size:20px;color:var(--mute);max-width:640px;margin:26px 0 34px;line-height:1.6}\n#tj .byline{display:flex;gap:22px;flex-wrap:wrap;font-size:13px;color:var(--mute)}\n#tj .hero-stats{display:grid;grid-template-columns:repeat(3,1fr);gap:0;margin-top:52px;border-top:1px solid var(--ink)}\n#tj .hero-stats div{padding:22px 24px 6px 0;border-right:1px solid var(--line)}\n#tj .hero-stats div+div{padding-left:24px}\n#tj .hero-stats div:last-child{border-right:0}\n#tj .hero-stats b{display:block;font-family:var(--serif);font-size:clamp(34px,5vw,58px);line-height:1;letter-spacing:-2px}\n#tj .hero-stats span{font-size:13px;color:var(--mute);display:block;margin-top:8px}\n#tj article{max-width:740px;padding-bottom:40px}\n#tj .lede p{margin:0 0 20px}\n#tj .lede{padding:44px 0 8px}\n#tj .pull{font-family:var(--serif);font-size:26px;line-height:1.35;border-left:3px solid var(--gold);padding-left:22px;margin:30px 0;font-style:italic}\n#tj section.arg{padding:60px 0 20px;border-top:1px solid var(--line);margin-top:40px}\n#tj .num{font-family:var(--serif);font-size:15px;letter-spacing:.14em;color:var(--gold);font-weight:700;text-transform:uppercase}\n#tj h2{font-family:var(--serif);font-size:clamp(28px,3.6vw,40px);line-height:1.15;letter-spacing:-1px;margin:10px 0 18px}\n#tj h3{font-size:12px;letter-spacing:.16em;text-transform:uppercase;color:var(--pu);margin:30px 0 8px}\n#tj p{margin-bottom:18px}\n#tj .src{font-size:13px;color:var(--mute);line-height:1.6}\n#tj .src a{color:var(--pu)}\n#tj .proof{background:var(--card);border:1px solid var(--line);border-radius:10px;padding:28px;margin:26px 0;box-shadow:0 18px 40px -26px rgba(40,34,80,.35)}\n#tj .proof .tag{font-size:11px;letter-spacing:.16em;text-transform:uppercase;color:var(--pu);font-weight:700;margin-bottom:6px}\n#tj .proof h4{font-family:var(--serif);font-size:21px;line-height:1.3;margin-bottom:18px}\n#tj .bar{display:grid;grid-template-columns:130px 1fr 72px;align-items:center;gap:12px;margin:10px 0;font-size:14px}\n#tj .bar .track{background:var(--lil);height:30px;border-radius:4px;overflow:hidden}\n#tj .bar .fill{height:100%;border-radius:4px;background:var(--ink);width:var(--w);transform-origin:left;animation:grow 1.1s cubic-bezier(.16,1,.3,1) both}\n#tj .bar .fill.pu{background:var(--pu)}\n#tj .bar .fill.gold{background:var(--gold)}\n#tj .bar .fill.coral{background:var(--coral)}\n#tj .bar b{font-family:var(--serif);font-size:20px;text-align:right}\n@keyframes grow{from{transform:scaleX(0)}to{transform:scaleX(1)}}\n#tj .note{font-size:12.5px;color:var(--mute);margin-top:14px;line-height:1.6;border-top:1px dashed var(--line);padding-top:12px}\n#tj .vs{display:grid;grid-template-columns:1fr 1fr;gap:12px}\n#tj .vs div{padding:22px;border-radius:8px;background:var(--lil)}\n#tj .vs div.d{background:var(--ink);color:#fff}\n#tj .vs small{font-size:12px;letter-spacing:.1em;text-transform:uppercase;opacity:.75;display:block}\n#tj .vs b{font-family:var(--serif);font-size:44px;letter-spacing:-1.5px;display:block;line-height:1.2}\n#tj .vs span{font-size:13px;opacity:.8}\n#tj .specs{display:grid;grid-template-columns:repeat(2,1fr);gap:12px}\n#tj .specs div{border:1px solid var(--line);border-top:3px solid var(--pu);border-radius:6px;padding:14px 16px;background:#fff}\n#tj .specs small{font-size:11px;letter-spacing:.12em;text-transform:uppercase;color:var(--pu);font-weight:700}\n#tj .specs b{display:block;font-family:var(--serif);font-size:24px;line-height:1.25;margin:2px 0}\n#tj .specs span{font-size:13px;color:var(--mute)}\n#tj .dark{background:var(--ink);color:#f3f1f8;margin:60px calc(50% - 50vw);padding:72px 0}\n#tj .dark .in{max-width:1120px;margin:0 auto;padding:0 28px}\n#tj .dark h2{color:#fff}\n#tj .dark .num{color:var(--gold)}\n#tj .pick{display:grid;grid-template-columns:1fr 1fr;gap:18px;margin-top:30px}\n#tj .pick div{border:1px solid rgba(255,255,255,.18);border-radius:10px;padding:26px}\n#tj .pick h4{font-family:var(--serif);font-size:24px;margin-bottom:12px}\n#tj .pick li{margin:0 0 8px 18px;font-size:15.5px;color:#d9d6e6}\n#tj .pick .t{border-color:var(--gold)}\n#tj .cta{margin:44px 0 0;display:flex;gap:14px;flex-wrap:wrap}\n#tj .cta .btn{background:#fff;color:var(--ink)}\n#tj .cta .btn:hover{background:var(--gold);color:#fff}\n#tj .faq details{border-top:1px solid var(--line);padding:16px 0}\n#tj .faq summary{font-weight:700;cursor:pointer;list-style:none}\n#tj .faq summary::after{content:\"+\";float:right;color:var(--pu)}\n#tj .faq details[open] summary::after{content:\"\u2013\"}\n#tj .faq p{margin:10px 0 0;color:var(--mute);font-size:15.5px}\n@media(max-width:900px){#tj .hero-stats{grid-template-columns:1fr}#tj .hero-stats div,#tj .hero-stats div+div{padding:16px 0;border-right:0;border-bottom:1px solid var(--line)}#tj .pick,#tj .vs,#tj .specs{grid-template-columns:1fr}#tj .bar{grid-template-columns:96px 1fr 60px;font-size:13px}#tj .proof{padding:20px}}\n#tj .hero h1{letter-spacing:-1px;line-height:1.18;font-size:clamp(34px,5.4vw,64px)}\n#tj .hero h1 em{font-style:normal}\n#tj .hero-stats b{letter-spacing:-1px}\n#tj h2{letter-spacing:0;line-height:1.3}\n#tj h3{letter-spacing:.08em}\n#tj .num,#tj .kicker,#tj .proof .tag,#tj .toc h4,#tj .specs small,#tj .vs small{letter-spacing:.08em}\n#tj .pull{font-style:normal;font-weight:700;line-height:1.5}\n#tj .vs b{letter-spacing:-1px}\n#tj table.ref{width:100%;border-collapse:collapse;font-size:15px}\n#tj table.ref th,#tj table.ref td{padding:12px 10px;border-bottom:1px solid var(--line);text-align:right}\n#tj table.ref th:first-child,#tj table.ref td:first-child{text-align:left;color:var(--mute)}\n#tj table.ref th{font-family:var(--serif);font-size:16px;vertical-align:bottom}\n#tj table.ref th small{font:400 11px var(--sans);color:var(--mute)}\n#tj table.ref td{font-family:var(--serif);font-size:18px}\n#tj table.ref td:first-child{font-family:var(--sans);font-size:14px}\n#tj .pick{grid-template-columns:repeat(2,1fr)}\n@media(max-width:900px){#tj table.ref th,#tj table.ref td{padding:10px 4px;font-size:14px}#tj table.ref td{font-size:15px}#tj .proof{overflow-x:auto}}\n#tj{font-size:17px;line-height:1.85;font-family:var(--sans)}\n#tj h2,#tj h3,#tj h4,#tj b,#tj p,#tj li,#tj td,#tj th,#tj summary{font-family:inherit}\n#tj h2,#tj .hero-stats b,#tj .vs b,#tj .specs b,#tj .bar b,#tj .proof h4,#tj .pick h4,#tj .pull,#tj table.ref th,#tj table.ref td,#tj .num{font-family:var(--serif)}\n#tj table.ref td:first-child{font-family:var(--sans)}\n#tj h2{font-size:clamp(26px,3.4vw,36px);color:var(--ink);font-weight:700}\n#tj h3{font-weight:700}\n#tj .hero{padding:8px 0 40px}\n#tj .hero .dek{margin:0 0 22px}\n#tj .hero-stats{margin-top:32px}\n#tj .tocrow{display:flex;flex-wrap:wrap;gap:8px 18px;font-size:13px;margin:26px 0 0;padding-top:14px;border-top:1px solid var(--line)}\n#tj .tocrow a{color:var(--mute);text-decoration:none;border-bottom:2px solid var(--line);padding-bottom:2px}\n#tj .tocrow a:hover{color:var(--pu);border-color:var(--pu)}\n#tj .lede{padding:28px 0 8px}\n#tj .dark{margin:56px 0;padding:56px 0;border-radius:12px}\n#tj .dark .in{padding:0 36px}\n#tj .pick ul{list-style:disc;margin:0;padding:0}\n#tj .faq summary::-webkit-details-marker{display:none}\n#tj .btn{color:#fff}\n#tj .dark h4,#tj .dark .pick h4{color:#fff}\n#tj table.ref,#tj table.ref tr,#tj table.ref th,#tj table.ref td{background:transparent!important;border-left:0!important;border-right:0!important;border-top:0!important}\n#tj table.ref th,#tj table.ref td{border-bottom:1px solid var(--line)!important}\n#tj table.ref thead th{border-bottom:1px solid var(--ink)!important;color:var(--ink)}\n#tj .cta .btn{color:var(--ink)}\n@media(max-width:700px){#tj .dark .in{padding:0 20px}#tj .hero-stats{grid-template-columns:1fr}#tj .hero-stats div,#tj .hero-stats div+div{padding:16px 0;border-right:0;border-bottom:1px solid var(--line)}#tj .pick,#tj .vs,#tj .specs{grid-template-columns:1fr}#tj .bar{grid-template-columns:96px 1fr 60px;font-size:13px}#tj .proof{padding:20px;overflow-x:auto}#tj table.ref th,#tj table.ref td{padding:10px 4px;font-size:14px}}\n\n<\/style>\n<div id=\"tj\">\n<section class=\"hero\">\n  <div class=\"kicker\"><i><\/i>AI Model Explainer \u00b7 Updated November 4, 2026<\/div>\n  <p class=\"dek\">OpenAI released <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> on October 26, positioned as a model for getting everyday work done at lower cost and higher speed. Using official figures and independent benchmarks, this article covers three things: what it costs, how fast it runs, and how often it gets facts wrong.<\/p>\n  <div class=\"byline\"><span>7 min read<\/span><span>Sources checked November 4, 2026<\/span><\/div>\n  <div class=\"hero-stats\">\n    <div><b>$1 \/ $5<\/b><span>API list price per 1M input \/ output tokens<\/span><\/div>\n    <div><b>16.7%<\/b><span>Factual error rate on OpenAI&#8217;s own test (low effort); the previous Terra scored 53.6%<\/span><\/div>\n    <div><b>1.05M<\/b><span>Token context window, with up to 128K tokens of output per request<\/span><\/div>\n  <\/div>\n  <nav class=\"tocrow\" aria-label=\"In this guide\"><a href=\"#cost\">1 \u00b7 Price<\/a><a href=\"#speed\">2 \u00b7 Speed<\/a><a href=\"#truth\">3 \u00b7 Accuracy<\/a><a href=\"#screen\">4 \u00b7 Screen and software<\/a><a href=\"#table\">5 \u00b7 Side by side<\/a><a href=\"#pick\">Four questions<\/a><a href=\"#faq\">\uc790\uc8fc \ubb3b\ub294 \uc9c8\ubb38<\/a><\/nav>\n<\/section>\n\n<div class=\"lede\">\n<p>Before you start a new project, do you check how much of your AI quota is left? If so, a new model sold on efficiency is worth a few minutes of your time, if only to see what its maker actually published.<\/p>\n<p>OpenAI released <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> on October 26, 2026, replacing GPT-5.6 Terra. OpenAI positions it as a model for getting everyday work done at lower cost and higher speed.<\/p>\n<p>That positioning covers three things: <strong>cost, speed and accuracy<\/strong>. We take them one at a time: the key point first, then a plain-language explanation, then the evidence. At the end there is a summary table and four questions people commonly ask when evaluating a model.<\/p>\n<div class=\"pull\">Price, speed and accuracy decide what each request costs, how long you wait, and whether you can rely on the answer.<\/div>\n<\/div>\n\n<section class=\"arg\" id=\"cost\">\n  <div class=\"num\">01 \u00b7 Cost<\/div>\n  <h2>Price: $1 in, $5 out per million tokens<\/h2>\n  <h3>Key point<\/h3>\n  <p>The standard API price for <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> is $1 per million input tokens and $5 per million output tokens. OpenAI says this is 50% lower than both <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-5.6 Terra<\/span> and GPT-6.1 Sol.<\/p>\n  <h3>In plain terms<\/h3>\n  <p>A token is roughly three quarters of an English word. Everything you send to the model and every answer it returns is billed in tokens. A typical task reads far more than it writes, so both the input price and the output price matter. Take one common job size: give the model 1 million tokens of material and have it write 100,000 tokens back.<\/p>\n  <div class=\"proof\">\n    <div class=\"tag\">Worked example \u00b7 price math<\/div>\n    <h4>Same job: 1M tokens in, 100K tokens out<\/h4>\n    <div class=\"vs\"><div><small>Previous-generation price (derived from the 50% figure)<\/small><b>$3.00<\/b><span>$2.00 input + $1.00 output<\/span><\/div><div class=\"d\"><small><span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span><\/small><b>$1.50<\/b><span>$1.00 input + $0.50 output<\/span><\/div><\/div>\n    <div class=\"note\">The previous-generation price is back-calculated from OpenAI&#8217;s &#8220;50% lower&#8221; statement and is not an official list price. The example assumes both generations use the same number of tokens; in real projects the same task can consume a different number of tokens. API pricing and subscription allowances are two separate systems.<\/div>\n  <\/div>\n  <h3>The evidence<\/h3>\n  <p class=\"src\">For subscribers, OpenAI says that the same usage allowance in <span class=\"notranslate\" translate=\"no\" data-no-translation>ChatGPT Work<\/span> \uadf8\ub9ac\uace0 <span class=\"notranslate\" translate=\"no\" data-no-translation>Codex<\/span> covers more tasks with Terra than with <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-5.6 Terra<\/span>, and substantially more than with Sol or Astra. It has not published an exact multiple. Sources: <a href=\"https:\/\/openai.com\/index\/introducing-gpt-6-1-terra\/\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI launch post<\/a>, <a href=\"https:\/\/developers.openai.com\/api\/docs\/models\/gpt-6.1-terra\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI API model page<\/a>.<\/p>\n<\/section>\n\n<section class=\"arg\" id=\"speed\">\n  <div class=\"num\">02 \u00b7 Speed<\/div>\n  <h2>Speed: OpenAI compared it with its own larger models<\/h2>\n  <h3>Key point<\/h3>\n  <p>OpenAI says Terra completes tasks more than three times faster than <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6 Astra<\/span> \uadf8\ub9ac\uace0 <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Sol<\/span>, and that it performs on par with Astra in OpenAI&#8217;s tests at one tenth of Astra&#8217;s price per token.<\/p>\n  <h3>In plain terms<\/h3>\n  <p>The comparison here is with the same company&#8217;s larger models, which are built to think longer and naturally take more time. Speed also depends on the reasoning effort setting: Terra offers six levels, from none to max. The higher the level, the longer the model thinks before it answers.<\/p>\n  <h3>The evidence<\/h3>\n  <p>Independent benchmarking firm <span class=\"notranslate\" translate=\"no\" data-no-translation>Artificial Analysis<\/span> measured the following at Terra&#8217;s highest reasoning setting.<\/p>\n  <div class=\"proof\">\n    <div class=\"tag\">Independent measurement \u00b7 Max effort<\/div>\n    <h4>Measured speed of <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span><\/h4>\n    <div class=\"specs\">\n      <div><small>Output speed<\/small><b>78 tokens\/s<\/b><span>How fast text is generated once the answer starts<\/span><\/div>\n      <div><small>Time to first token<\/small><b>141 s<\/b><span>At the highest setting, including thinking time<\/span><\/div>\n    <\/div>\n    <div class=\"note\">\uc6d0\ucc9c: <a href=\"https:\/\/artificialanalysis.ai\/models\/comparisons\/gpt-6-1-terra-vs-claude-opus-5-5-xhigh\" rel=\"nofollow noopener\" target=\"_blank\"><span class=\"notranslate\" translate=\"no\" data-no-translation>Artificial Analysis<\/span> comparison<\/a>, Terra (Max), retrieved November 4, 2026. Lower reasoning settings produce the first token sooner. Speeds change with provider load and time of day; the same page showed 91 tokens\/s and 144 s on October 29, 2026.<\/div>\n  <\/div>\n<\/section>\n\n<section class=\"arg\" id=\"truth\">\n  <div class=\"num\">03 \u00b7 Accuracy<\/div>\n  <h2>Accuracy: fewer factual errors than the previous generation<\/h2>\n  <h3>Key point<\/h3>\n  <p>On OpenAI&#8217;s internal factual-error test, <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span>&#8216;s error rate at low reasoning effort fell from 53.6% to 16.7%. At maximum effort it fell to 4.3%, close to the larger Astra and Sol models.<\/p>\n  <h3>In plain terms<\/h3>\n  <p>A &#8220;factual error&#8221; here means the model gave a confident answer that was wrong. These are the errors most likely to slip into a report, because nobody checks every line. The test is made of questions chosen to trip models up, so the absolute numbers look high. What is worth reading is the direction and the size of the drop.<\/p>\n  <h3>The evidence<\/h3>\n  <div class=\"proof\">\n    <div class=\"tag\">OpenAI internal test<\/div>\n    <h4>Factual error rate at low reasoning effort (lower is better)<\/h4>\n    <div class=\"bar\"><span><span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-5.6 Terra<\/span><\/span><div class=\"track\"><div class=\"fill coral\" style=\"--w:53.6%\"><\/div><\/div><b>53.6%<\/b><\/div>\n    <div class=\"bar\"><span><span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span><\/span><div class=\"track\"><div class=\"fill pu\" style=\"--w:16.7%\"><\/div><\/div><b>16.7%<\/b><\/div>\n    <div class=\"note\">Source: OpenAI&#8217;s published results on its own question set (the 4.3% maximum-effort figure is as reported by <span class=\"notranslate\" translate=\"no\" data-no-translation>BleepingComputer<\/span>, citing OpenAI). These results come from one specific question set and are not the same as an everyday error rate. For facts that matter, check the original source.<\/div>\n  <\/div>\n<\/section>\n\n<section class=\"arg\" id=\"screen\">\n  <div class=\"num\">04 \u00b7 Capabilities<\/div>\n  <h2>Capabilities: it can read a screen and operate software<\/h2>\n  <h3>Key point<\/h3>\n  <p>Besides text, Terra accepts images and can understand what is on a computer screen. On platforms that provide the right tools, it can click, type and switch between apps to complete multi-step work.<\/p>\n  <h3>In plain terms<\/h3>\n  <p>Before, you could ask a model how a product page should be designed. Now you can hand it the product photos and specs and ask it to try building a page that opens, then check the result yourself. Going from giving advice to doing the work is what &#8220;computer use&#8221; means here.<\/p>\n  <h3>The evidence<\/h3>\n  <p>OpenAI&#8217;s published specs show what it can take in and what it can produce. Note the limits: no audio or video input, and text-only output.<\/p>\n  <div class=\"proof\">\n    <div class=\"tag\">Official specs<\/div>\n    <h4><span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> at a glance<\/h4>\n    <div class=\"specs\">\n      <div><small>Context window<\/small><b>1.05M tokens<\/b><span>Room for a whole shelf of long novels in one request<\/span><\/div>\n      <div><small>Max output<\/small><b>128K tokens<\/b><span>Long documents in a single response<\/span><\/div>\n      <div><small>Input \/ output<\/small><b>Text + images \u2192 text<\/b><span>No audio or video input<\/span><\/div>\n      <div><small>Knowledge cutoff<\/small><b>June 1, 2026<\/b><span>Six reasoning levels, from none to max<\/span><\/div>\n    <\/div>\n    <div class=\"note\">\uc6d0\ucc9c: <a href=\"https:\/\/developers.openai.com\/api\/docs\/models\/gpt-6.1-terra\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI API model page<\/a>, retrieved November 4, 2026. Requests with more than 272K input tokens are billed at 2x the input and cache rates and 1.5x the output rate for the whole request. Computer-use ability depends on the tools and permissions the host platform provides.<\/div>\n  <\/div>\n<\/section>\n\n<section class=\"arg\" id=\"table\">\n  <div class=\"num\">05 \u00b7 Side by side<\/div>\n  <h2>The published numbers in one place<\/h2>\n  <h3>Key point<\/h3>\n  <p>The table below puts together the list prices of two models and the figures one independent benchmarking firm measured for both with the same method.<\/p>\n  <div class=\"proof\">\n    <div class=\"tag\">List prices and independent measurements<\/div>\n    <h4>Published numbers for <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> \uadf8\ub9ac\uace0 <span class=\"notranslate\" translate=\"no\" data-no-translation>Claude Opus 5.5<\/span><\/h4>\n    <table class=\"ref\">\n      <thead><tr><th>\ubbf8\ud130\ubc95<\/th><th><span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span><br><small>Max effort<\/small><\/th><th><span class=\"notranslate\" translate=\"no\" data-no-translation>Claude Opus 5.5<\/span><br><small>xhigh effort<\/small><\/th><\/tr><\/thead>\n      <tbody>\n        <tr><td>Price per 1M input tokens<\/td><td>$1.00<\/td><td>$4.00<\/td><\/tr>\n        <tr><td>Price per 1M output tokens<\/td><td>$5.00<\/td><td>$20.00<\/td><\/tr>\n        <tr><td>Intelligence Index<\/td><td>51<\/td><td>57<\/td><\/tr>\n        <tr><td>Cost to run the full index<\/td><td>$972<\/td><td>$4,766<\/td><\/tr>\n        <tr><td>Output speed (tokens\/s)<\/td><td>78<\/td><td>107<\/td><\/tr>\n        <tr><td>Time to first token (s)<\/td><td>141.07<\/td><td>77.81<\/td><\/tr>\n        <tr><td>Context window<\/td><td>1.05M tokens<\/td><td>1M tokens<\/td><\/tr>\n      <\/tbody>\n    <\/table>\n    <div class=\"note\">Sources: <a href=\"https:\/\/artificialanalysis.ai\/models\/comparisons\/gpt-6-1-terra-vs-claude-opus-5-5-xhigh\" rel=\"nofollow noopener\" target=\"_blank\"><span class=\"notranslate\" translate=\"no\" data-no-translation>Artificial Analysis<\/span> comparison<\/a> (index, cost, speed, time to first token), and the <a href=\"https:\/\/developers.openai.com\/api\/docs\/models\/gpt-6.1-terra\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI API model page<\/a> \uadf8\ub9ac\uace0 <a href=\"https:\/\/platform.claude.com\/docs\/en\/about-claude\/pricing\" rel=\"nofollow noopener\" target=\"_blank\">Anthropic pricing page<\/a> (prices, context windows), retrieved November 4, 2026. The two models were released close together, and both were tested at a high reasoning setting. Other aggregators report different index values because they use different scoring versions.<\/div>\n  <\/div>\n  <h3>Reading the table<\/h3>\n  <ul style=\"margin:0 0 18px 20px\">\n    <li>Prices are per token. The total bill also depends on how many tokens a model uses to finish the job.<\/li>\n    <li>The Intelligence Index summarizes many tests. It reflects average performance, not how a model handles your specific task.<\/li>\n    <li>Speed and time to first token vary with reasoning setting, provider load and time of day.<\/li>\n  <\/ul>\n  <p>For how we read another recent model launch, see our <a href=\"https:\/\/imastudio.com\/ko\/blog\/deepseek-v4-review\" style=\"color:var(--pu)\">DeepSeek V4 review<\/a>. For the image and video models we cover, see our <a href=\"https:\/\/imastudio.com\/ko\/blog\/ai-model-guides-hub\" style=\"color:var(--pu)\">AI \ubaa8\ub378 \uac00\uc774\ub4dc<\/a>.<\/p>\n<\/section>\n\n<div class=\"dark\" id=\"pick\"><div class=\"in\">\n  <div class=\"num\">Four questions<\/div>\n  <h2>Four things to weigh when evaluating a model<\/h2>\n  <div class=\"pick\">\n    <div><h4>1 \u00b7 How often does the task run?<\/h4><ul><li>The more runs, the more a difference in unit price adds up<\/li><li>For occasional tasks, price may not be the main factor<\/li><\/ul><\/div>\n    <div><h4>2 \u00b7 What does one error cost?<\/h4><ul><li>Is a person or script checking the output?<\/li><li>Are errors easy to spot, or do they slip quietly downstream?<\/li><\/ul><\/div>\n    <div><h4>3 \u00b7 How much reasoning does it need?<\/h4><ul><li>Reasoning effort is adjustable; higher settings mean longer waits<\/li><li>Everyday sorting and rewriting need less than hard multi-step problems<\/li><\/ul><\/div>\n    <div><h4>4 \u00b7 What form are the inputs and outputs?<\/h4><ul><li>Terra takes text and images and writes text<\/li><li>Audio or video work needs extra tools<\/li><\/ul><\/div>\n  <\/div>\n  <p style=\"margin-top:26px;color:#bdb9cf;font-size:15px\">The most direct way to tell whether a model fits your work is to try it on a real task you already know the answer to.<\/p>\n  <div class=\"cta\"><a class=\"btn\" href=\"https:\/\/imastudio.com\/ko\">See What You Can Create on <span class=\"notranslate\" translate=\"no\" data-no-translation>IMA Studio<\/span><\/a><\/div>\n<\/div><\/div>\n\n<section class=\"faq\" id=\"faq\" style=\"padding:20px 0 40px\">\n  <div class=\"num\">\uc790\uc8fc \ubb3b\ub294 \uc9c8\ubb38<\/div><h2>Quick answers<\/h2>\n  <details><summary>When was <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span> released?<\/summary><p>October 26, 2026. It is available in <span class=\"notranslate\" translate=\"no\" data-no-translation>ChatGPT Work<\/span> \uadf8\ub9ac\uace0 <span class=\"notranslate\" translate=\"no\" data-no-translation>Codex<\/span> on the Plus, Pro, Enterprise and Edu plans. The API model name is gpt-6.1-terra.<\/p><\/details>\n  <details><summary>How should I read the API price?<\/summary><p>You pay per token, with separate prices for input and output. A project&#8217;s total cost depends on how much you give the model, how much you ask it to write, and how many tokens it actually uses to finish.<\/p><\/details>\n  <details><summary>Are price and answer quality the same thing?<\/summary><p>No. Price and quality are separate dimensions, and each needs its own test results.<\/p><\/details>\n  <details><summary>Can Terra work with video or audio?<\/summary><p>No. Input is text and images only, and output is text only. For video work, extract key frames first or use tools provided by the platform.<\/p><\/details>\n  <details><summary>Can I use Terra on <span class=\"notranslate\" translate=\"no\" data-no-translation>IMA Studio<\/span>?<\/summary><p>As of publication, <span class=\"notranslate\" translate=\"no\" data-no-translation>IMA Studio<\/span> does not offer GPT-6.1 Terra.<\/p><\/details>\n<\/section>\n<p class=\"src\" style=\"padding-bottom:30px\">Sources, all retrieved November 4, 2026: <a href=\"https:\/\/openai.com\/index\/introducing-gpt-6-1-terra\/\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI: Introducing <span class=\"notranslate\" translate=\"no\" data-no-translation>GPT-6.1 Terra<\/span><\/a> \u00b7 <a href=\"https:\/\/developers.openai.com\/api\/docs\/models\/gpt-6.1-terra\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI API model page<\/a> \u00b7 <a href=\"https:\/\/platform.claude.com\/docs\/en\/about-claude\/pricing\" rel=\"nofollow noopener\" target=\"_blank\">Anthropic pricing page<\/a> \u00b7 <a href=\"https:\/\/artificialanalysis.ai\/models\/comparisons\/gpt-6-1-terra-vs-claude-opus-5-5-xhigh\" rel=\"nofollow noopener\" target=\"_blank\"><span class=\"notranslate\" translate=\"no\" data-no-translation>Artificial Analysis<\/span> comparison<\/a>. Figures labeled &#8220;OpenAI internal test&#8221; are the vendor&#8217;s results on its own question set.<\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>What OpenAI and independent testers published about GPT-6.1 Terra: API price, speed, factual-error rate and specs, with the source and the caveat behind each number.<\/p>","protected":false},"author":26,"featured_media":15549,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"rank_math_title":"GPT-6.1 Terra: Price, Speed and Accuracy Numbers Explained","rank_math_description":"GPT-6.1 Terra costs $1\/$5 per million tokens. See OpenAI's published speed and factual-error figures, independent benchmarks and specs, with sources.","footnotes":""},"categories":[49],"tags":[51],"class_list":["post-15546","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tools-models","tag-type-guides"],"_links":{"self":[{"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/posts\/15546","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/users\/26"}],"replies":[{"embeddable":true,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/comments?post=15546"}],"version-history":[{"count":2,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/posts\/15546\/revisions"}],"predecessor-version":[{"id":15548,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/posts\/15546\/revisions\/15548"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/media\/15549"}],"wp:attachment":[{"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/media?parent=15546"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/categories?post=15546"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/imastudio.com\/ko\/wp-json\/wp\/v2\/tags?post=15546"}],"curies":[{"name":"\uc6cc\ub4dc\ud504\ub808\uc2a4","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}