OpenAI lança o GPT-5.6 (Sol, Terra, Luna): uma família de modelos em três níveis com chamada programática de ferramentas na Responses API

OpenAI just moved the GPT-5.6 family to general availability, following a limited preview. The release ships three models rather than one. Sol is the flagship, Terra is the balanced everyday tier, and Luna is the most cost-efficient.

• GPT-5.6 ships three tiers — Sol, Terra, Luna — priced from $1/$6 to $5/$30 per 1M tokens.

• Sol leads the Artificial Analysis Coding Agent Index at 80, 2.8 points above Claude Fable 5.

• Programmatic Tool Calling runs model-written JavaScript in an isolated V8 runtime with no network access.

• ultra runs four agents in parallel, lifting Terminal-Bench 2.1 from 88.8% to 91.9%.

• SWE-Bench Pro remains a gap: Sol’s 64.6% trails Claude Mythos 5’s 80.3% by roughly 15 points.

What is GPT-5.6?

Three models, one generation, priced per 1M tokens. Sol is $5 input and $30 output. Terra is $2.50 and $15. Luna is $1 and $6.

Availability differs by surface:

• Chat: Plus, Pro, Business, and Enterprise users access Sol at medium and higher effort. Pro and Enterprise can also select GPT-5.6 Sol Pro.

• ChatGPT Work and Codex: Free and Go users access Terra. Paid users choose among all three and set effort per model. max is available to all users with GPT-5.6 access and is toggled in settings.

• API: All three tiers are available. Programmatic Tool Calling and a multi-agent beta both live in the Responses API.

Prompt caching also changed. GPT-5.6 supports explicit cache breakpoints and a 30-minute minimum cache life. Cache writes are billed at 1.25x the model’s uncached input rate. Cache reads continue to receive the 90% cached-input discount.

Performance

Furthermore, Agents’ Last Exam evaluates long-running professional workflows across 55 fields. OpenAI reports a new high of 53.6 for Sol. It describes this as eclipsing Claude Fable 5 (adaptive reasoning) by 13.1 points.

OpenAI’s own eval table lists Sol at 52.7% and Fable 5 at 40.5%. The 13.1-point gap matches 53.6 minus 40.5, so the Fable 5 baseline is consistent across both. Only Sol’s figure differs. OpenAI does not label which reasoning configuration produced 53.6.

On the Artificial Analysis Coding Agent Index v1.1, Sol at max reasoning scores 80. That is 2.8 points above Fable 5. OpenAI reports it does so using less than half the output tokens and less than half the time.

Sol sets new state-of-the-art results on Terminal-Bench 2.1 and DeepSWE. It reaches 92.2% on BrowseComp and 62.6% on OSWorld 2.0. On OSWorld it surpasses Claude Opus 4.8 while using 85% fewer output tokens.

EvalGPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.5Claude Fable 5Claude Opus 4.8Gemini 3.1 Pro PreviewAA Coding Agent Index v1.18077.474.676.477.272.542.7AA Intelligence Index v4.158.95551.254.859.955.746.5Terminal-Bench 2.188.8%87.4%84.7%85.6%83.1%78.9%70.7%DeepSWE v1.172.7%69.6%67.2%67%69.7%59%11.8%SWE-Bench Pro64.6%63.4%62.7%59.4%80%69.2%54.2%Agents’ Last Exam52.7%50.4%50.3%46.9%40.5%45.2%32.1%GDPval-AA v2 (Elo)1,747.81,5931,591.81,493.71,759.61,600.1962.3BrowseComp90.4%87.5%83.3%84.4%—84.3%85.9%OSWorld 2.062.6%50.2%45.6%47.5%—54.8%—Toolathlon58%53.1%53.4%55.6%61.7%59.9%48.8%Source: OpenAI’s published GPT-5.6 eval tables. Sol Ultra reaches 91.9% on Terminal-Bench 2.1 and 92.2% on BrowseComp. Claude Mythos 5 scores 80.3% on SWE-Bench Pro and 88% on Terminal-Bench 2.1. A dash means the score was not reported.

Where GPT-5.6 Does Not Lead

However, four gaps are worth naming:

• SWE-Bench Pro: Sol scores 64.6%. Claude Mythos 5 scores 80.3% and Fable 5 scores 80%. That is a roughly 15-point deficit on a widely watched coding eval.

• Broad intelligence and knowledge work: Fable 5 leads the Artificial Analysis Intelligence Index v4.1, 59.9 to 58.9. Fable 5 also leads GDPval-AA v2 by about 12 Elo. On HealthBench Professional, Fable 5 scores 60.9% against Sol’s 60.5%.

• Tool use: On Toolathlon, Sol scores 58%. Fable 5 reaches 61.7% and Opus 4.8 reaches 59.9%. Luna also edges out Terra here, inverting the tier order.

• Long context: Luna drops to 41.3% on OpenAI MRCR v2 8-needle, at both 256K–512K and 512K–1M. Sol scores 73.8% at 512K–1M, slightly below GPT-5.5’s 74%.

Interactive Explainer

GPT-5.6 Explorer — Marktechpost

#mtp-gpt56-root *,#mtp-gpt56-root *::before,#mtp-gpt56-root *::after{box-sizing:border-box!important}

html,body{margin:0!important;padding:0!important;background:transparent!important}

#mtp-gpt56-root{

--paper:#F7F7F4; --card:#FFFFFF; --ink:#101418; --ink-2:#4A5058;

--rule:#DEDEd6; --rule-2:#EBEBE4;

--pine:#1F6F5C; --pine-soft:#E4F0EC;

--clay:#B4531E; --clay-soft:#FAEBE0;

--slate:#3A4A63; --amber:#8A6D1F;

--mono:"JetBrains Mono",ui-monospace,SFMono-Regular,Menlo,monospace;

--serif:"IBM Plex Serif",Georgia,serif;

--sans:"Inter",system-ui,-apple-system,Segoe UI,sans-serif;

font-family:var(--sans)!important; color:var(--ink)!important;

background:var(--paper)!important; border:1px solid var(--rule)!important;

border-radius:4px!important; overflow:hidden!important; max-width:100%!important;

line-height:1.5!important; -webkit-font-smoothing:antialiased;

#mtp-gpt56-root p,#mtp-gpt56-root h1,#mtp-gpt56-root h2,#mtp-gpt56-root h3,

#mtp-gpt56-root h4,#mtp-gpt56-root ul,#mtp-gpt56-root li{margin:0!important;padding:0!important;list-style:none!important}

#mtp-gpt56-root button{font-family:inherit!important;cursor:pointer!important;border:0;background:none}

#mtp-gpt56-root button:focus-visible,#mtp-gpt56-root select:focus-visible,

#mtp-gpt56-root input:focus-visible{outline:2px solid var(--pine)!important;outline-offset:2px!important}

/* header */

#mtp-gpt56-root .m-head{padding:18px 22px 14px!important;border-bottom:1px solid var(--rule)!important;background:var(--card)!important}

#mtp-gpt56-root .m-eyebrow{font-family:var(--mono)!important;font-size:10px!important;letter-spacing:.14em!important;

text-transform:uppercase!important;color:var(--pine)!important;font-weight:700!important}

#mtp-gpt56-root .m-title{font-family:var(--serif)!important;font-size:23px!important;font-weight:600!important;

letter-spacing:-.01em!important;margin-top:6px!important;line-height:1.2!important}

#mtp-gpt56-root .m-sub{font-size:13px!important;color:var(--ink-2)!important;margin-top:6px!important;max-width:62ch!important}

/* stepper */

#mtp-gpt56-root .m-steps{display:flex!important;gap:0!important;border-bottom:1px solid var(--rule)!important;

background:var(--card)!important;overflow-x:auto!important;-webkit-overflow-scrolling:touch}

#mtp-gpt56-root .m-step{flex:1 1 0!important;min-width:132px!important;padding:11px 12px!important;text-align:left!important;

border-right:1px solid var(--rule-2)!important;color:var(--ink-2)!important;font-size:12px!important;

font-weight:500!important;white-space:nowrap!important;transition:background .15s,color .15s}

#mtp-gpt56-root .m-step:last-child{border-right:0!important}

#mtp-gpt56-root .m-step .m-num{font-family:var(--mono)!important;font-size:10px!important;color:var(--rule)!important;

display:block!important;margin-bottom:3px!important;font-weight:700!important}

#mtp-gpt56-root .m-step:hover{background:var(--paper)!important}

#mtp-gpt56-root .m-step.on{color:var(--ink)!important;background:var(--pine-soft)!important;box-shadow:inset 0 -2px 0 var(--pine)!important}

#mtp-gpt56-root .m-step.on .m-num{color:var(--pine)!important}

/* stage */

#mtp-gpt56-root .m-stage{padding:22px!important;background:var(--paper)!important;min-height:420px!important}

#mtp-gpt56-root .m-panel{display:none!important}

#mtp-gpt56-root .m-panel.on{display:block!important}

#mtp-gpt56-root .m-h{font-family:var(--serif)!important;font-size:17px!important;font-weight:600!important;margin-bottom:4px!important}

#mtp-gpt56-root .m-note{font-size:12.5px!important;color:var(--ink-2)!important;margin-bottom:16px!important;max-width:70ch!important}

/* tier cards */

#mtp-gpt56-root .m-tiers{display:grid!important;grid-template-columns:repeat(3,1fr)!important;gap:12px!important}

#mtp-gpt56-root .m-tier{background:var(--card)!important;border:1px solid var(--rule)!important;border-radius:3px!important;

padding:14px!important;text-align:left!important;transition:border-color .15s,transform .12s;width:100%!important}

#mtp-gpt56-root .m-tier:hover{transform:translateY(-1px)}

#mtp-gpt56-root .m-tier.on{border-color:var(--pine)!important;box-shadow:0 0 0 1px var(--pine) inset!important}

#mtp-gpt56-root .m-tier-name{font-family:var(--mono)!important;font-size:13px!important;font-weight:700!important}

#mtp-gpt56-root .m-tier-role{font-size:11px!important;color:var(--ink-2)!important;margin-top:3px!important;white-space:normal!important}

#mtp-gpt56-root .m-tier-price{font-family:var(--mono)!important;font-size:11px!important;margin-top:10px!important;

padding-top:9px!important;border-top:1px dashed var(--rule)!important;color:var(--clay)!important;font-weight:500!important}

#mtp-gpt56-root .m-tier-id{font-family:var(--mono)!important;font-size:10px!important;color:var(--ink-2)!important;margin-top:5px!important}

#mtp-gpt56-root .m-fact{background:var(--card)!important;border:1px solid var(--rule)!important;border-left:3px solid var(--pine)!important;

padding:12px 14px!important;margin-top:14px!important;font-size:12.5px!important;color:var(--ink-2)!important;border-radius:2px!important}

#mtp-gpt56-root .m-fact b{color:var(--ink)!important;font-weight:600!important}

/* controls */

#mtp-gpt56-root .m-grid2{display:grid!important;grid-template-columns:1fr 1fr!important;gap:20px!important;align-items:start!important}

#mtp-gpt56-root .m-field{margin-bottom:15px!important}

#mtp-gpt56-root .m-label{display:flex!important;justify-content:space-between!important;align-items:baseline!important;

font-size:11.5px!important;color:var(--ink-2)!important;margin-bottom:6px!important;font-weight:500!important}

#mtp-gpt56-root .m-val{font-family:var(--mono)!important;color:var(--ink)!important;font-weight:700!important;font-size:12px!important}

#mtp-gpt56-root input[type=range]{width:100%!important;-webkit-appearance:none;appearance:none;height:3px!important;

background:var(--rule)!important;border-radius:2px!important;margin:0!important}

#mtp-gpt56-root input[type=range]::-webkit-slider-thumb{-webkit-appearance:none;width:15px;height:15px;border-radius:50%;

background:var(--pine);cursor:pointer;border:2px solid #fff;box-shadow:0 0 0 1px var(--pine)}

#mtp-gpt56-root input[type=range]::-moz-range-thumb{width:13px;height:13px;border-radius:50%;background:var(--pine);

cursor:pointer;border:2px solid #fff}

#mtp-gpt56-root select{width:100%!important;padding:8px 10px!important;border:1px solid var(--rule)!important;

border-radius:3px!important;background:var(--card)!important;font-size:12.5px!important;color:var(--ink)!important;font-family:var(--sans)!important}

#mtp-gpt56-root .m-toggle{display:flex!important;align-items:center!important;gap:8px!important;font-size:12px!important;

color:var(--ink-2)!important;cursor:pointer!important;user-select:none}

#mtp-gpt56-root .m-toggle input{accent-color:var(--pine);width:14px;height:14px;cursor:pointer}

/* cost readout — signature */

#mtp-gpt56-root .m-readout{background:var(--card)!important;border:1px solid var(--rule)!important;border-radius:3px!important;padding:0!important;overflow:hidden!important}

#mtp-gpt56-root .m-readout-h{padding:9px 14px!important;border-bottom:1px solid var(--rule-2)!important;

font-family:var(--mono)!important;font-size:10px!important;letter-spacing:.12em!important;text-transform:uppercase!important;color:var(--ink-2)!important}

#mtp-gpt56-root .m-row{display:flex!important;align-items:center!important;gap:10px!important;padding:11px 14px!important;

border-bottom:1px solid var(--rule-2)!important}

#mtp-gpt56-root .m-row:last-child{border-bottom:0!important}

#mtp-gpt56-root .m-row-n{font-family:var(--mono)!important;font-size:11.5px!important;font-weight:700!important;width:52px!important;flex:none!important}

#mtp-gpt56-root .m-bar-wrap{flex:1 1 auto!important;height:6px!important;background:var(--rule-2)!important;border-radius:3px!important;overflow:hidden!important}

#mtp-gpt56-root .m-bar{height:100%!important;background:var(--pine)!important;width:0%!important;transition:width .35s cubic-bezier(.4,0,.2,1)!important}

#mtp-gpt56-root .m-row.hi .m-bar{background:var(--clay)!important}

#mtp-gpt56-root .m-row-v{font-family:var(--mono)!important;font-size:12.5px!important;font-weight:700!important;

width:96px!important;text-align:right!important;flex:none!important}

#mtp-gpt56-root .m-row-x{font-family:var(--mono)!important;font-size:10.5px!important;color:var(--ink-2)!important;width:62px!important;text-align:right!important;flex:none!important}

/* bench chart */

#mtp-gpt56-root .m-brow{display:flex!important;align-items:center!important;gap:10px!important;margin-bottom:7px!important}

#mtp-gpt56-root .m-bname{font-family:var(--mono)!important;font-size:10.5px!important;width:132px!important;flex:none!important;

color:var(--ink-2)!important;white-space:nowrap!important;overflow:hidden!important;text-overflow:ellipsis!important}

#mtp-gpt56-root .m-btrack{flex:1 1 auto!important;height:17px!important;background:var(--rule-2)!important;border-radius:2px!important;position:relative!important;overflow:hidden!important}

#mtp-gpt56-root .m-bfill{height:100%!important;width:0%!important;border-radius:2px!important;transition:width .5s cubic-bezier(.4,0,.2,1)!important}

#mtp-gpt56-root .m-bval{font-family:var(--mono)!important;font-size:11px!important;font-weight:700!important;width:58px!important;text-align:right!important;flex:none!important}

#mtp-gpt56-root .m-legend{display:flex!important;flex-wrap:wrap!important;gap:14px!important;margin-top:14px!important;

padding-top:12px!important;border-top:1px dashed var(--rule)!important;font-size:11px!important;color:var(--ink-2)!important}

#mtp-gpt56-root .m-dot{display:inline-block!important;width:9px!important;height:9px!important;border-radius:2px!important;margin-right:5px!important;vertical-align:-1px!important}

/* ultra */

#mtp-gpt56-root .m-agents{display:flex!important;gap:7px!important;margin:14px 0 4px!important;flex-wrap:wrap!important}

#mtp-gpt56-root .m-agent{width:44px!important;height:44px!important;border:1px solid var(--rule)!important;border-radius:3px!important;

background:var(--card)!important;display:flex!important;align-items:center!important;justify-content:center!important;

font-family:var(--mono)!important;font-size:10px!important;color:var(--rule)!important;transition:all .3s}

#mtp-gpt56-root .m-agent.live{border-color:var(--pine)!important;background:var(--pine-soft)!important;color:var(--pine)!important;font-weight:700!important}

#mtp-gpt56-root .m-switch{display:inline-flex!important;border:1px solid var(--rule)!important;border-radius:3px!important;overflow:hidden!important;background:var(--card)!important}

#mtp-gpt56-root .m-switch button{padding:8px 15px!important;font-size:12px!important;font-weight:500!important;color:var(--ink-2)!important;

border-right:1px solid var(--rule)!important}

#mtp-gpt56-root .m-switch button:last-child{border-right:0!important}

#mtp-gpt56-root .m-switch button.on{background:var(--pine)!important;color:#fff!important;font-weight:600!important}

/* nav + footer */

#mtp-gpt56-root .m-nav{display:flex!important;justify-content:space-between!important;align-items:center!important;

padding:12px 22px!important;border-top:1px solid var(--rule)!important;background:var(--card)!important}

#mtp-gpt56-root .m-nav button{padding:8px 16px!important;border:1px solid var(--rule)!important;border-radius:3px!important;

font-size:12px!important;font-weight:500!important;color:var(--ink)!important;background:var(--card)!important}

#mtp-gpt56-root .m-nav button:hover:not(:disabled){border-color:var(--pine)!important;color:var(--pine)!important}

#mtp-gpt56-root .m-nav button:disabled{opacity:.35!important;cursor:not-allowed!important}

#mtp-gpt56-root .m-dots{display:flex!important;gap:6px!important}

#mtp-gpt56-root .m-dotn{width:7px!important;height:7px!important;border-radius:50%!important;background:var(--rule)!important;padding:0!important}

#mtp-gpt56-root .m-dotn.on{background:var(--pine)!important;width:18px!important;border-radius:4px!important}

#mtp-gpt56-root .m-foot{padding:11px 22px!important;background:var(--paper)!important;border-top:1px solid var(--rule)!important;

display:flex!important;justify-content:space-between!important;align-items:center!important;gap:10px!important;flex-wrap:wrap!important}

#mtp-gpt56-root .m-foot span{font-family:var(--mono)!important;font-size:10px!important;color:var(--ink-2)!important;letter-spacing:.04em!important}

#mtp-gpt56-root .m-foot b{color:var(--pine)!important;font-weight:700!important}

#mtp-gpt56-root hr,#mtp-gpt56-root p:empty,#mtp-gpt56-root del,#mtp-gpt56-root s{display:none!important}

@media (max-width:640px){

#mtp-gpt56-root .m-head{padding:14px 15px 12px!important}

#mtp-gpt56-root .m-title{font-size:19px!important}

#mtp-gpt56-root .m-stage{padding:15px!important;min-height:0!important}

#mtp-gpt56-root .m-grid2{grid-template-columns:1fr!important;gap:14px!important}

#mtp-gpt56-root .m-tiers{grid-template-columns:1fr!important}

#mtp-gpt56-root .m-step{min-width:118px!important;font-size:11px!important;padding:9px 10px!important}

#mtp-gpt56-root .m-bname{width:88px!important;font-size:9.5px!important}

#mtp-gpt56-root .m-row-v{width:80px!important;font-size:11.5px!important}

#mtp-gpt56-root .m-row-x{width:48px!important}

#mtp-gpt56-root .m-nav{padding:10px 15px!important}

#mtp-gpt56-root .m-foot{padding:10px 15px!important}

@media (prefers-reduced-motion:reduce){#mtp-gpt56-root *{transition:none!important}}

Interactive · OpenAI GPT‑5.6 · July 9, 2026

GPT‑5.6 tier, cost and benchmark explorer

Every number below is taken from OpenAI’s published GPT‑5.6 eval tables and price list. Move the controls to see how tier choice changes spend and score.

01Model tiers

02Cost calculator

03Benchmarks

04Ultra mode

Three durable capability tiers

The number is the generation. Sol, Terra and Luna are tiers that advance on their own cadence. Select a tier to see its role and rate card.

What a workload actually costs

Set your per-request token volume. Costs use OpenAI’s published per-1M rates. Cached input reads keep the 90% cached-input discount.

Input tokens per request12,000

Output tokens per request3,000

Requests per day2,000

Share of input served from cache0%

Estimated monthly spend · 30 days

Cache billing: from GPT‑5.6 onward, cache writes bill at 1.25x the uncached input rate. Cache reads keep the 90% discount, with a 30‑minute minimum cache life and explicit cache breakpoints.

Benchmark scores, side by side

Values are OpenAI’s published eval table. A dash in that table means the score was not reported, so the model is omitted here. OpenAI states its latency and cost figures are simulated offline, not measured in production.

GPT‑5.6 family

Ultra: four agents, by default

Ultra coordinates four agents in parallel by default. It trades higher token use for a stronger score and faster time‑to‑result. Toggle to compare against Sol’s one‑agent baseline. OpenAI also charts 16‑agent runs on BrowseComp and SEC‑Bench Pro.

Sol · default

Sol Ultra · 4 agents

Score · single agent

Terminal-Bench 2.1

SEC-Bench Pro

Where to get it: ultra runs in ChatGPT Work for Pro and Enterprise, and in Codex for Plus and higher. In the API, the multi‑agent beta in the Responses API builds ultra‑like flows.

Source: OpenAI, “GPT‑5.6” · verified July 9, 2026

Built by MARKTECHPOST

(function(){

var root = document.getElementById('mtp-gpt56-root');

/* ---------- data (OpenAI published tables) ---------- */

var TIERS = [

{k:'sol', n:'Sol', id:'see API models page', role:'Flagship. Hardest coding, agents, security research.', pin:5, pout:30,

fact:'Sol scores 88.8% on Terminal-Bench 2.1 and 62.6% on OSWorld 2.0. It is the only tier with an ultra setting. Pro and Enterprise chat users can also select GPT‑5.6 Sol Pro.'},

{k:'terra', n:'Terra', id:'see API models page', role:'Balanced default. High-volume business work.', pin:2.5, pout:15,

fact:'Terra is positioned as competitive with GPT-5.5 at half the price. It scores 87.4% on Terminal-Bench 2.1 against GPT-5.5’s 85.6%. Free and Go users get Terra in ChatGPT Work and Codex.'},

{k:'luna', n:'Luna', id:'see API models page', role:'Fastest and cheapest. Summarize, draft, classify.', pin:1, pout:6,

fact:'Luna reaches 62.7% on SWE-Bench Pro against GPT-5.5’s 59.4%, at one-fifth of Sol’s output rate. It trails GPT-5.5 on Terminal-Bench 2.1 (84.7% vs 85.6%). Long-context is its weak spot: 41.3% on MRCR v2 8-needle.'}

var VENDOR = {oai56:'#1F6F5C', oai55:'#8A6D1F', anth:'#3A4A63', goog:'#B4531E'};

var BENCH = [

{n:'Artificial Analysis Coding Agent Index v1.1', max:100, note:'Sol at max reasoning sets a new state of the art at 80, which is 2.8 points above Claude Fable 5, using less than half the output tokens.',

r:[['GPT-5.6 Sol',80,'oai56'],['GPT-5.6 Terra',77.4,'oai56'],['Claude Fable 5',77.2,'anth'],['GPT-5.5',76.4,'oai55'],['GPT-5.6 Luna',74.6,'oai56'],['Claude Opus 4.8',72.5,'anth'],['Gemini 3.1 Pro Preview',42.7,'goog']]},

{n:'Terminal-Bench 2.1', max:100, note:'Complex command-line workflows. Sol Ultra reaches 91.9%. Terra (87.4%) clears GPT-5.5, and even Luna (84.7%) sits above Claude Fable 5.',

r:[['GPT-5.6 Sol Ultra',91.9,'oai56'],['GPT-5.6 Sol',88.8,'oai56'],['Claude Mythos 5',88,'anth'],['GPT-5.6 Terra',87.4,'oai56'],['GPT-5.5',85.6,'oai55'],['GPT-5.6 Luna',84.7,'oai56'],['Claude Fable 5',83.1,'anth'],['Claude Opus 4.8',78.9,'anth'],['Gemini 3.1 Pro Preview',70.7,'goog']]},

{n:'SWE-Bench Pro', max:100, note:'This is where GPT-5.6 does not lead. Claude Mythos 5 (80.3%) and Fable 5 (80%) sit roughly 15 points above Sol. OpenAI does not headline this benchmark.',

r:[['Claude Mythos 5',80.3,'anth'],['Claude Fable 5',80,'anth'],['Claude Mythos Preview',77.8,'anth'],['Claude Opus 4.8',69.2,'anth'],['GPT-5.6 Sol',64.6,'oai56'],['GPT-5.6 Terra',63.4,'oai56'],['GPT-5.6 Luna',62.7,'oai56'],['GPT-5.5',59.4,'oai55'],['Gemini 3.1 Pro Preview',54.2,'goog']]},

{n:'DeepSWE v1.1', max:100, note:'Long-horizon engineering in real codebases. All three GPT-5.6 tiers clear GPT-5.5, and Sol takes the top score.',

r:[['GPT-5.6 Sol',72.7,'oai56'],['Claude Fable 5',69.7,'anth'],['GPT-5.6 Terra',69.6,'oai56'],['GPT-5.6 Luna',67.2,'oai56'],['GPT-5.5',67,'oai55'],['Claude Opus 4.8',59,'anth'],['Gemini 3.1 Pro Preview',11.8,'goog']]},

{n:'Agents’ Last Exam', max:100, note:'Long-running professional workflows across 55 fields. Even Luna (50.3%) clears Opus 4.8 and Fable 5 on this eval.',

r:[['GPT-5.6 Sol',52.7,'oai56'],['GPT-5.6 Terra',50.4,'oai56'],['GPT-5.6 Luna',50.3,'oai56'],['GPT-5.5',46.9,'oai55'],['Claude Opus 4.8',45.2,'anth'],['Claude Fable 5',40.5,'anth'],['Gemini 3.1 Pro Preview',32.1,'goog']]},

{n:'Artificial Analysis Intelligence Index v4.1', max:70, note:'Claude Fable 5 (59.9) still leads the broad intelligence index. Sol trails it by one point at roughly half the estimated cost.',

r:[['Claude Fable 5',59.9,'anth'],['GPT-5.6 Sol',58.9,'oai56'],['Claude Opus 4.8',55.7,'anth'],['GPT-5.6 Terra',55,'oai56'],['GPT-5.5',54.8,'oai55'],['GPT-5.6 Luna',51.2,'oai56'],['Gemini 3.5 Flash',50.2,'goog'],['Gemini 3.1 Pro Preview',46.5,'goog']]},

{n:'BrowseComp', max:100, note:'Agentic browsing. Sol Ultra sets a new state of the art at 92.2%.',

r:[['GPT-5.6 Sol Ultra',92.2,'oai56'],['GPT-5.6 Sol',90.4,'oai56'],['Claude Mythos 5',88,'anth'],['Claude Mythos Preview',87.9,'anth'],['GPT-5.6 Terra',87.5,'oai56'],['Gemini 3.1 Pro Preview',85.9,'goog'],['GPT-5.5',84.4,'oai55'],['Claude Opus 4.8',84.3,'anth'],['GPT-5.6 Luna',83.3,'oai56']]},

{n:'OSWorld 2.0', max:100, note:'Computer use. Sol reaches 62.6% and surpasses Opus 4.8 while using 85% fewer output tokens.',

r:[['GPT-5.6 Sol',62.6,'oai56'],['Claude Opus 4.8',54.8,'anth'],['GPT-5.6 Terra',50.2,'oai56'],['GPT-5.5',47.5,'oai55'],['GPT-5.6 Luna',45.6,'oai56']]},

{n:'ExploitBench', max:100, note:'Cyber capability, evaluated with reduced safeguards. Sol jumps from GPT-5.5’s 47.9% to 73.5%, but Claude Mythos 5 leads at 78%.',

r:[['Claude Mythos 5',78,'anth'],['Claude Mythos Preview',74.2,'anth'],['GPT-5.6 Sol',73.5,'oai56'],['GPT-5.6 Terra',52.9,'oai56'],['GPT-5.5',47.9,'oai55'],['Claude Opus 4.8',40,'anth'],['GPT-5.6 Luna',33.2,'oai56']]},

{n:'GDPval-AA v2 (Elo)', max:1800, note:'Professional knowledge work, scored in Elo. Claude Fable 5 (1,759.6) leads Sol (1,747.8) by roughly 12 Elo.',

r:[['Claude Fable 5',1759.6,'anth'],['GPT-5.6 Sol',1747.8,'oai56'],['Claude Opus 4.8',1600.1,'anth'],['GPT-5.6 Terra',1593,'oai56'],['GPT-5.6 Luna',1591.8,'oai56'],['GPT-5.5',1493.7,'oai55'],['Gemini 3.5 Flash',1348.8,'goog'],['Gemini 3.1 Pro Preview',962.3,'goog']]},

{n:'Toolathlon', max:100, note:'Tool use. Sol trails Fable 5 and Opus 4.8 here. Luna (53.4%) also edges out Terra (53.1%), inverting the tier order.',

r:[['Claude Fable 5',61.7,'anth'],['Claude Mythos 5',61.7,'anth'],['Claude Mythos Preview',61.1,'anth'],['Claude Opus 4.8',59.9,'anth'],['GPT-5.6 Sol',58,'oai56'],['GPT-5.5',55.6,'oai55'],['GPT-5.6 Luna',53.4,'oai56'],['GPT-5.6 Terra',53.1,'oai56'],['Gemini 3.1 Pro Preview',48.8,'goog']]}

var ULTRA = {

single:{tb:88.8, bc:90.4, sb:71.2},

quad: {tb:91.9, bc:92.2, sb:74.3}

/* ---------- slide machinery ---------- */

var idx = 0, N = 4;

var panels = root.querySelectorAll('.m-panel');

var steps = root.querySelectorAll('.m-step');

var dotsEl = document.getElementById('mDots');

var prevB = document.getElementById('mPrev');

var nextB = document.getElementById('mNext');

for(var d=0; d= 1000) return '$' + n.toLocaleString('en-US',{maximumFractionDigits:0});

return '$' + n.toLocaleString('en-US',{minimumFractionDigits:2, maximumFractionDigits:2});

function calc(){

var ti=+rIn.value, to=+rOut.value, rq=+rReq.value, ch=+rCache.value/100;

document.getElementById('lIn').textContent = fmt(ti);

document.getElementById('lOut').textContent = fmt(to);

document.getElementById('lReq').textContent = fmt(rq);

document.getElementById('lCache').textContent = (ch*100).toFixed(0) + '%';

var out = {};

TIERS.forEach(function(t){

var inCost = (ti*(1-ch)/1e6)*t.pin + (ti*ch/1e6)*(t.pin*0.10); // 90% cached-input discount

var outCost = (to/1e6)*t.pout;

out[t.k] = (inCost + outCost) * rq * 30;

var mx = out.sol || 1;

['sol','terra','luna'].forEach(function(k){

var cap = k.charAt(0).toUpperCase()+k.slice(1);

document.getElementById('b'+cap).style.width = Math.max(2,(out[k]/mx)*100) + '%';

document.getElementById('v'+cap).textContent = money(out[k]);

document.getElementById('xTerra').textContent = (out.sol/out.terra).toFixed(1) + 'x less';

document.getElementById('xLuna').textContent = (out.sol/out.luna).toFixed(1) + 'x less';

[rIn,rOut,rReq,rCache].forEach(function(el){ el.addEventListener('input', calc); });

/* ---------- 3: benchmarks ---------- */

var sel = document.getElementById('mBench');

BENCH.forEach(function(b,i){

var o = document.createElement('option');

o.value = i; o.innerHTML = b.n;

sel.appendChild(o);

function drawBench(){

var b = BENCH[+sel.value || 0];

var wrap = document.getElementById('mBars');

wrap.innerHTML = '';

b.r.forEach(function(row){

var div = document.createElement('div');

div.className = 'm-brow';

div.innerHTML = ''+row[0]+'' +

''+row[1]+'';

wrap.appendChild(div);

var f = div.querySelector('.m-bfill');

setTimeout(function(){ f.style.width = ((row[1]/b.max)*100) + '%'; }, 30);

document.getElementById('mBenchNote').innerHTML = b.note;

sel.addEventListener('change', drawBench);

drawBench();

/* ---------- 4: ultra ---------- */

var uMode = 0;

var agentsEl = document.getElementById('mAgents');

for(var a=0;a 0 ? '+' + delta.toFixed(1) : '\u2014';

document.getElementById('mUltra').querySelectorAll('button').forEach(function(btn){

btn.onclick = function(){

document.getElementById('mUltra').querySelectorAll('button').forEach(function(x){x.classList.remove('on');});

btn.classList.add('on');

uMode = +btn.dataset.u;

drawUltra();

drawUltra();

/* ---------- auto-resize (component offsetHeight) ---------- */

function ping(){

var h = root.offsetHeight + 40;

if(window.parent && window.parent !== window){

window.parent.postMessage({mtpFrame:'gpt56-explorer', height:h}, '*');

}catch(e){}

window.addEventListener('load', function(){ ping(); setTimeout(ping,300); setTimeout(ping,900); });

window.addEventListener('resize', ping);

setInterval(ping, 1500);

Strengths and Weaknesses

• Sets the Artificial Analysis Coding Agent Index at 80, above Fable 5’s 77.2

• Programmatic Tool Calling has a documented API contract and named-customer token reductions of 38% to 63.5%

• All three tiers clear GPT-5.5 on DeepSWE v1.1, SWE-Bench Pro, and Agents’ Last Exam

• A 5x price spread across tiers lets teams route by task difficulty

• Explicit cache breakpoints and a 30-minute minimum cache life make caching predictable

• Sol reaches 62.6% on OSWorld 2.0 using 85% fewer output tokens than Opus 4.8

At the same time, the release faces a few distinct hurdles:

• SWE-Bench Pro trails Claude Mythos 5 and Fable 5 by roughly 15 points

• Fable 5 leads on AA Intelligence Index v4.1, GDPval-AA v2, HealthBench Professional, and Toolathlon

• The 53.6 Agents’ Last Exam headline does not appear in OpenAI’s own table, which lists 52.7%

• Luna scores 41.3% on MRCR v2 8-needle, and Luna trails GPT-5.5 on Terminal-Bench 2.1

• Cyber scores are measured with reduced safeguards; production behavior will differ

• Cache writes now cost 1.25x the uncached input rate, a new line item to model

• Latency and cost claims are OpenAI’s offline simulations, not measured production numbers

• GPT-5.6: Frontier intelligence that scales with your ambition — openai.com

• Previewing GPT-5.6 Sol: a next-generation model — openai.com

• GPT-5.6 system card — deploymentsafety.openai.com

• Programmatic Tool Calling — developers.openai.com

• Multi-agent — developers.openai.com

• Prompt caching — developers.openai.com

• Agents’ Last Exam — agents-last-exam.org

• Artificial Analysis Intelligence Index — artificialanalysis.ai

The post OpenAI Releases GPT-5.6 (Sol, Terra, Luna): A Three-Tier Model Family With Programmatic Tool Calling in the Responses API appeared first on MarkTechPost.