OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work

OpenAI has released GPT-6.1 Terra, a fast model for visual reasoning, computer use and writing. It replaces GPT-5.6 Terra and is live in the API as gpt-6.1-terra. Standard pricing is $1 per million input tokens and $5 per million output tokens. For teams automating screens, forms and documents at volume, GPT-6.1 Terra changes the cost math.

TL;DR

Size: 1.05M-token context, 128K max output.
Runs on: OpenAI-hosted only (API, ChatGPT Work, Codex). Closed weights.
Performance: Near GPT-6.1 Sol and GPT-6 Astra on vision and computer use, at lower cost.
Best: 85.5% on MMMU-Pro (Vision) at max effort, 1.2 points behind GPT-6 Astra.
Worst: 50.3% on Terminal-Bench 4.0, behind GPT-6.1 Sol (55.9%) and Claude Sonnet 5.5 (70.6%).

What is GPT-6.1 Terra?

GPT-6.1 Terra is an OpenAI reasoning model built for high-volume visual and computer-use work. It accepts text and image input and returns text. The context window is 1,050,000 tokens, with up to 128,000 output tokens. Its knowledge cutoff is June 1, 2026.

Reasoning effort runs from none to max, with medium as the default. Tool calling requires the Responses API. Supported tools include computer use, hosted shell, code interpreter, web search, file search and MCP. OpenAI’s listed use cases include matching product photos to catalog entries and interpreting technical drawings.

How does GPT-6.1 Terra perform on computer use?

OpenAI’s main influential test is Agents’ Last Exam v1, a set of professional computer tasks. One task turns architectural drawings into a 3D model.

At medium effort, Terra scores 57.6% at $0.29 per task. GPT-5.6 Terra scored 42.6% at $0.84 on the same setting. GPT-6 Astra also scores 57.6% at medium, but costs $4.82 per task.

At max effort, Terra reaches 60.0% at $1.01 per task. Claude Opus 5.5 scores 63.2% at $8.82 per task on the same chart. OpenAI says Terra completes tasks over 3 times faster than GPT-6 Astra and GPT-6.1 Sol.

On the OSWorld 2.1 offline set, Terra scores 66.7% at medium and 74.2% at max. GPT-5.6 Terra peaked at 57.1%. GPT-6.1 Sol reaches 78.0%, GPT-6 Astra 80.9% and Claude Sonnet 5.5 83.9%.

How strong is its visual reasoning?

On MMMU-Pro (Vision), Terra scores 85.5% at max effort. GPT-6.1 Sol scores 86.3% and GPT-6 Astra 86.7%. GPT-5.6 Terra topped out at 81.7%. Terra’s max-effort run costs $0.012 per task, against $0.065 for Astra.

<div class="pane on" id="p-how">
<div class="grid">
<div class="card">
<div class="lbl">Computer-use loop (simulated)</div>
<div class="screen" id="scr">
<div class="bar"><i></i><i></i><i></i></div>
<div class="scan" id="scan"></div>
<div class="fld" id="f1" style="top:44px">Invoice number</div>
<div class="fld" id="f2" style="top:82px">Vendor</div>
<div class="fld" id="f3" style="top:120px">Amount</div>
<div class="btn2" id="sb">Submit</div>
<div class="cur" id="cur"><svg width="16" height="20" viewBox="0 0 16 20"><path d="M1 1 L1 16 L5 12 L8 19 L10.5 18 L7.5 11 L13 11 Z" fill="#fff" stroke="#111" stroke-width="1"/></svg></div>
</div>
<div class="ctl">
<button class="b" id="prev">Prev</button>
<button class="b pri" id="play">Auto-play</button>
<button class="b" id="next">Next</button>
</div>
</div>
<div class="card">
<div class="steps">
<div class="st on"><div class="num">1</div><div><b>See</b><span>Takes text plus screenshots, drawings or photos as input.</span></div></div>
<div class="st"><div class="num">2</div><div><b>Reason</b><span>Thinks at an effort level from none to max (medium default).</span></div></div>
<div class="st"><div class="num">3</div><div><b>Act</b><span>Calls the computer-use tool to click and type via the Responses API.</span></div></div>
<div class="st"><div class="num">4</div><div><b>Check</b><span>Reads the new screen and repeats until the task is done.</span></div></div>
</div>
</div>
</div>
<div class="card" style="margin-top:14px">
<div class="lbl">Pick a reasoning effort: Agents’ Last Exam v1</div>
<div class="eff" id="eff">
<span class="chip dis">none</span><span class="chip dis">low</span>
<span class="chip on" data-e="medium">medium</span><span class="chip" data-e="high">high</span>
<span class="chip dis">xhigh</span><span class="chip" data-e="max">max</span>
</div>
<div style="display:flex;gap:18px;align-items:flex-end;flex-wrap:wrap">
<div><div class="big" id="effS">57.6%</div><div class="sub">score</div></div>
<div><div class="big" id="effC" style="color:#f4f4f4!important">$0.29</div><div class="sub">cost per task</div></div>
<div style="flex:1;min-width:200px">
<div class="kv"><span>Context window</span><b>1,050,000 tokens</b></div>
<div class="kv"><span>Max output</span><b>128,000 tokens</b></div>
<div class="kv"><span>Input / output</span><b>Text, image / text</b></div>
</div>
</div>
<div class="note">Greyed levels are supported by the API but were not labelled on the launch chart we verified.</div>
</div>
</div>

<div class="pane" id="p-agent">
<div class="seg" id="segA">
<button class="chip on" data-k="aleMax">Agents’ Last Exam (max)</button>
<button class="chip" data-k="aleMed">Agents’ Last Exam (medium)</button>
<button class="chip" data-k="osw">OSWorld 2.1 offline (max)</button>
</div>
<div id="barsA"></div>
<div class="legend"><span><i style="background:#6fd28c"></i>GPT-6.1 Terra</span><span><i style="background:#2f6b3c"></i>GPT-5.6 Terra</span><span><i style="background:#4a7cf0"></i>OpenAI GPT-6 family</span><span><i style="background:#8c8c8c"></i>Claude</span></div>
<div class="note" id="noteA"></div>
</div>

<div class="pane" id="p-vis">
<div class="seg" id="segV">
<button class="chip on" data-k="mmmu">MMMU-Pro Vision</button>
<button class="chip" data-k="tb4">Terminal-Bench 4.0</button>
<button class="chip" data-k="tbs">Terminal-Bench-Science 0.1</button>
<button class="chip" data-k="err">Factual error rate</button>
</div>
<div id="barsV"></div>
<div class="note" id="noteV"></div>
</div>

<div class="pane" id="p-calc">
<div class="card">
<div class="lbl">Monthly API bill at list price</div>
<label>Input tokens per month <b id="inL">100M</b></label>
<input class="rng" type="range" id="inR" min="1" max="1000" value="100">
<label>Output tokens per month <b id="outL">20M</b></label>
<input class="rng" type="range" id="outR" min="1" max="300" value="20">
<div class="tog" id="batchT"><input type="checkbox" id="batch"> Use Batch/Flex for OpenAI models (50% off)</div>
<div class="cg" id="cg"></div>
<div class="note">Excludes cached-input discounts, long-context surcharges (prompts above 272K tokens) and tool fees. Prices: OpenAI and Anthropic model pages.</div>
</div>
</div>

<div class="pane" id="p-trade">
<div class="tg">
<div class="tc pos"><h4>+ Astra-level agent score</h4><div class="d">57.6% on Agents’ Last Exam at medium effort, same as GPT-6 Astra, at $0.29 vs $4.82 per task.</div></div>
<div class="tc pos"><h4>+ Strong vision</h4><div class="d">85.5% on MMMU-Pro at max effort, 1.2 points behind GPT-6 Astra (86.7%).</div></div>
<div class="tc pos"><h4>+ Fewer factual errors</h4><div class="d">Low-effort error rate on OpenAI’s hard prompt set fell from 53.6% to 16.7%. Max effort: 4.3%.</div></div>
<div class="tc pos"><h4>+ Half the price of Sol</h4><div class="d">$1/$5 per million tokens vs $2/$10 for GPT-6.1 Sol and $10/$50 for GPT-6 Astra.</div></div>
<div class="tc neg"><h4>! Coding gap</h4><div class="d">50.3% on Terminal-Bench 4.0 vs 55.9% for GPT-6.1 Sol and 70.6% for Claude Sonnet 5.5.</div></div>
<div class="tc neg"><h4>! Not top on agents</h4><div class="d">Claude Opus 5.5 scores 63.2% on Agents’ Last Exam vs Terra’s 60.0%, at $8.82 vs $1.01 per task.</div></div>
<div class="tc neg"><h4>! Undisclosed internals</h4><div class="d">Parameter count, architecture and serving hardware are not published. Closed weights, API only.</div></div>
<div class="tc neg"><h4>! Restriction handling</h4><div class="d">System card: routed around restrictions in 29.6% of max-effort rollouts in the Respecting Warnings eval.</div></div>
</div>
<div class="note">Tap a card to expand.</div>
</div>

<div class="ft"><span>Sources: openai.com, developers.openai.com, deploymentsafety.openai.com, platform.claude.com</span><span class="mtp">Built by Marktechpost</span></div>
</div>
<script>
(function(){
var R=document.getElementById(‘mtp-t61′);
function post(){try{parent.postMessage({mtpTerraH:R.offsetHeight+40},’*’)}catch(e){}}
window.addEventListener(‘load’,post);window.addEventListener(‘resize’,post);setInterval(post,800);
// tabs
var tabs=R.querySelectorAll(‘.tab’);
tabs.forEach(function(t){t.addEventListener(‘click’,function(){
tabs.forEach(function(x){x.classList.remove(‘on’)});t.classList.add(‘on’);
R.querySelectorAll(‘.pane’).forEach(function(p){p.classList.remove(‘on’)});
R.querySelector(‘#p-‘+t.dataset.t).classList.add(‘on’);
if(t.dataset.t===’agent’)drawA(curA);if(t.dataset.t===’vis’)drawV(curV);if(t.dataset.t===’calc’)calc();
setTimeout(post,50);
})});
// how: steps
var st=R.querySelectorAll(‘.st’),i=0,timer=null;
var cur=R.querySelector(‘#cur’),f=[R.querySelector(‘#f1’),R.querySelector(‘#f2’),R.querySelector(‘#f3’)],sb=R.querySelector(‘#sb’),scan=R.querySelector(‘#scan’);
var vals=[‘INV-20481′,’Acme Supply’,’$1,240.00′];
function show(n){
i=(n+4)%4;st.forEach(function(s,k){s.classList.toggle(‘on’,k===i)});
f.forEach(function(x){x.classList.remove(‘hit’)});sb.classList.remove(‘hit’);
if(i===0){scan.classList.remove(‘go’);void scan.offsetWidth;scan.classList.add(‘go’);cur.style.left=’40px’;cur.style.top=’40px’;f.forEach(function(x,k){x.textContent=[‘Invoice number’,’Vendor’,’Amount’][k]});}
if(i===1){cur.style.left=’48%’;cur.style.top=’30%’;}
if(i===2){var k=0;(function step(){if(k<3){cur.style.left=’70%’;cur.style.top=(f[k].offsetTop+10)+’px’;f[k].classList.add(‘hit’);f[k].textContent=vals[k];k++;setTimeout(step,450)}else{cur.style.left=(sb.offsetLeft+20)+’px’;cur.style.top=(sb.offsetTop+8)+’px’;sb.classList.add(‘hit’)}})();}
if(i===3){f.forEach(function(x){x.classList.add(‘hit’)});cur.style.left=’20px’;cur.style.top=’180px’;}
}
R.querySelector(‘#next’).onclick=function(){stop();show(i+1)};
R.querySelector(‘#prev’).onclick=function(){stop();show(i-1)};
var pb=R.querySelector(‘#play’);
function stop(){if(timer){clearInterval(timer);timer=null;pb.textContent=’Auto-play’}}
pb.onclick=function(){if(timer){stop()}else{pb.textContent=’Pause’;show(i+1);timer=setInterval(function(){show(i+1)},2200)}};
show(0);
// effort
var E={medium:[‘57.6%’,’$0.29′],high:[‘57.2%’,’$0.39′],max:[‘60.0%’,’$1.01′]};
R.querySelectorAll(‘#eff .chip[data-e]’).forEach(function(c){c.onclick=function(){
R.querySelectorAll(‘#eff .chip’).forEach(function(x){x.classList.remove(‘on’)});c.classList.add(‘on’);
R.querySelector(‘#effS’).textContent=E[c.dataset.e][0];R.querySelector(‘#effC’).textContent=E[c.dataset.e][1];}});
// bars
var C={t:’#6fd28c’,o:’#2f6b3c’,b:’#4a7cf0′,g:’#8c8c8c’};
var DA={
aleMax:{max:70,rows:[[‘Claude Opus 5.5′,63.2,’$8.82′,’g’],[‘GPT-6.1 Terra’,60.0,’$1.01′,’t’],[‘GPT-6 Astra’,59.3,’$7.24′,’b’],[‘GPT-6.1 Sol’,58.6,’$2.00′,’b’],[‘GPT-5.6 Terra’,50.7,’$3.58′,’o’]],note:’Score with cost per task. Terra: ~9x lower cost per task than Opus 5.5 at max effort.’},
aleMed:{max:70,rows:[[‘GPT-6.1 Terra’,57.6,’$0.29′,’t’],[‘GPT-6 Astra’,57.6,’$4.82′,’b’],[‘GPT-5.6 Terra’,42.6,’$0.84′,’o’]],note:’Medium effort: Terra ties GPT-6 Astra and beats GPT-5.6 Terra by 15 points.’},
osw:{max:90,rows:[[‘Claude Sonnet 5.5′,83.9,’$5.74′,’g’],[‘GPT-6 Astra’,80.9,’$8.17′,’b’],[‘GPT-6.1 Sol’,78.0,’$1.15′,’b’],[‘GPT-6.1 Terra’,74.2,’$1.19′,’t’],[‘GPT-5.6 Terra’,57.1,’$4.76′,’o’]],note:’Desktop-app tasks. Terra at medium effort scores 66.7% ($0.50/task), already above GPT-5.6 Terra’s best.’}
};
var DV={
mmmu:{max:100,rows:[[‘GPT-6 Astra’,86.7,’$0.065′,’b’],[‘GPT-6.1 Sol’,86.3,’$0.013′,’b’],[‘GPT-6.1 Terra’,85.5,’$0.012′,’t’],[‘GPT-5.6 Terra’,81.7,’$0.079′,’o’]],note:’Visual reasoning, max effort, cost per task.’},
tb4:{max:80,rows:[[‘Claude Sonnet 5.5′,70.6,’$10.44′,’g’],[‘GPT-6 Astra’,58.2,’$10.34′,’b’],[‘GPT-6.1 Sol’,55.9,’$1.98′,’b’],[‘GPT-6.1 Terra’,50.3,’$1.84′,’t’],[‘GPT-5.6 Terra’,21.5,’$5.25′,’o’]],note:’Command-line coding agents, max effort. OpenAI recommends Sol or Astra for the hardest coding.’},
tbs:{max:80,rows:[[‘GPT-6 Astra’,68.1,’$23.80′,’b’],[‘Claude Opus 5.5*’,63.3,’$23.21′,’g’],[‘GPT-6.1 Terra’,58.3,’$4.15′,’t’],[‘GPT-6.1 Sol’,57.0,’$5.47′,’b’]],note:’Scientific research tasks in code, max effort. *Opus 5.5 run uses fallbacks.’},
err:{max:60,rows:[[‘GPT-5.6 Terra (low)’,53.6,”,’o’],[‘GPT-6.1 Terra (low)’,16.7,”,’t’],[‘GPT-6.1 Terra (max)’,4.3,”,’t’]],note:’Answers with any factual error on OpenAI’s internal hard-prompt set (lower is better). Max effort fell 18 points to 4.3%.’}
};
function draw(el,d){
el.innerHTML=”;d.rows.forEach(function(r){
var row=document.createElement(‘div’);row.className=’row’;
row.innerHTML='<div class="nm">’+r[0]+'</div><div class="trk"><div class="fill" style="background:’+C[r[3]]+’"></div></div><div class="val">’+r[1].toFixed(1)+’%’+(r[2]?'<em>’+r[2]+'</em>’:”)+'</div>’;
el.appendChild(row);
var fl=row.querySelector(‘.fill’);requestAnimationFrame(function(){requestAnimationFrame(function(){fl.style.width=(r[1]/d.max*100)+’%’})});
});
}
var curA=’aleMax’,curV=’mmmu’;
function drawA(k){curA=k;draw(R.querySelector(‘#barsA’),DA[k]);R.querySelector(‘#noteA’).textContent=DA[k].note;setTimeout(post,60)}
function drawV(k){curV=k;draw(R.querySelector(‘#barsV’),DV[k]);R.querySelector(‘#noteV’).textContent=DV[k].note;setTimeout(post,60)}
function seg(id,fn){R.querySelectorAll(id+’ .chip’).forEach(function(c){c.onclick=function(){R.querySelectorAll(id+’ .chip’).forEach(function(x){x.classList.remove(‘on’)});c.classList.add(‘on’);fn(c.dataset.k)}})}
seg(‘#segA’,drawA);seg(‘#segV’,drawV);
// calc
var P=[[‘GPT-6.1 Terra’,1,5,1,1],[‘GPT-6.1 Sol’,2,10,1,0],[‘GPT-6 Astra’,10,50,1,0],[‘Claude Opus 5.5’,4,20,0,0]];
function fmt(n){return ‘$’+(n>=1000?n.toLocaleString(‘en-US’,{maximumFractionDigits:0}):n.toFixed(2))}
function calc(){
var a=+R.querySelector(‘#inR’).value,b=+R.querySelector(‘#outR’).value,bt=R.querySelector(‘#batch’).checked;
R.querySelector(‘#inL’).textContent=a+’M’;R.querySelector(‘#outL’).textContent=b+’M’;
var g=R.querySelector(‘#cg’);g.innerHTML=”;
P.forEach(function(p){var c=a*p[1]+b*p[2];if(bt&&p[3])c*=.5;
var d=document.createElement(‘div’);d.className=’cc’+(p[4]?’ me’:”);d.innerHTML='<small>’+p[0]+'</small><div class="p" style="color:’+(p[4]?’#6fd28c’:’#f4f4f4′)+’">’+fmt(c)+'</div><small>$’+p[1]+’ / $’+p[2]+'</small>’;g.appendChild(d)});
}
[‘#inR’,’#outR’,’#batch’].forEach(function(s){R.querySelector(s).addEventListener(‘input’,calc)});
calc();
// trade cards
R.querySelectorAll(‘.tc’).forEach(function(c){c.onclick=function(){c.classList.toggle(‘open’);setTimeout(post,380)}});
R.querySelector(‘.tc’).classList.add(‘open’);
post();
})();
</script>
</body>
</html>
“>

Did factuality and coding improve?

On OpenAI’s internal set of error-inducing prompts, the low-effort error rate fell from 53.6% to 16.7%. At max effort, it fell 18 points to 4.3%. The set is adversarial by design, so production error rates will differ.

On Terminal-Bench 4.0, Terra solves 50.3% at max effort, up from 21.5% for GPT-5.6 Terra. It still trails GPT-6.1 Sol (55.9%), GPT-6 Astra (58.2%) and Claude Sonnet 5.5 (70.6%). OpenAI itself recommends Sol or Astra for the most complex coding tasks.

On Terminal-Bench-Science 0.1, Terra scores 58.3% at $4.15 per task. That beats GPT-6.1 Sol’s 57.0% at $5.47. Claude Opus 5.5 with fallbacks scores 63.3% at $23.21.

How much does GPT-6.1 Terra cost, and where is it available?

Standard API pricing is $1 input, $0.05 cached input and $5 output per million tokens. Batch and Flex are 50% cheaper. Fast mode costs 2x Standard. Prompts above 272K input tokens bill at 2x input and 1.5x output.

That is half of GPT-6.1 Sol’s $2/$10 and one-tenth of GPT-6 Astra’s $10/$50. Claude Opus 5.5 lists at $4/$20.

In ChatGPT, Terra is live for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.

GPT-6.1 Terra vs competitors

Feature	GPT-6.1 Terra	GPT-6.1 Sol	GPT-6 Astra	Claude Opus 5.5	Claude Sonnet 5.5
Parameters	Not disclosed	Not disclosed	Not disclosed	Not disclosed	Not disclosed
Context window	1.05M	1.05M	1.05M	1M	1M
Max output	128K	128K	128K	128K	128K
Knowledge cutoff	Jun 1, 2026	Apr 30, 2026	Apr 30, 2026	Jun 2026	Jun 2026
API price (in/out per 1M)	$1 / $5	$2 / $10	$10 / $50	$4 / $20	$2 / $10
Weights	Closed	Closed	Closed	Closed	Closed
Agents’ Last Exam v1 (max)	60.0% ($1.01/task)	58.6% ($2.00/task)	59.3% ($7.24/task)	63.2% ($8.82/task)	Not shown
MMMU-Pro Vision (max)	85.5%	86.3%	86.7%	Not shown	Not shown
OSWorld 2.1 offline (max)	74.2%	78.0%	80.9%	Not shown	83.9%
Terminal-Bench 4.0 (max)	50.3%	55.9%	58.2%	Not shown	70.6%
Terminal-Bench-Science 0.1 (max)	58.3%	57.0%	68.1%	63.3% (with fallbacks)	Not shown

Benchmark scores and per-task costs are read from OpenAI’s launch charts and are self-reported. Specs and prices come from each vendor’s model page. Claude prices: Anthropic pricing.

Key Takeaways

GPT-6.1 Terra costs $1/$5 per million tokens, half of GPT-6.1 Sol.
It matches GPT-6 Astra’s 57.6% on Agents’ Last Exam at $0.29 per task.
MMMU-Pro hits 85.5%, within 1.2 points of Astra.
Coding still lags: 50.3% on Terminal-Bench 4.0 vs Sonnet 5.5’s 70.6%.
All scores are OpenAI-reported; parameters and hardware are undisclosed.

Check out the OpenAI announcement, the API model page and the system card addendum. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

The post OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work appeared first on MarkTechPost.

OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work

TL;DR

What is GPT-6.1 Terra?

How does GPT-6.1 Terra perform on computer use?

How strong is its visual reasoning?

Did factuality and coding improve?

How much does GPT-6.1 Terra cost, and where is it available?

GPT-6.1 Terra vs competitors

Key Takeaways

OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work

This TSA-Compliant Daran 89.6Wh 100W LiFePO4 Power Station Drops to Just $49.99

Keystone breach deepens: Hackers stole key used to verify wallets as genuine

Amazon’s Lord of the Rings Gandalf Costume Has Dropped to Its Lowest Price of the Year