Is it deployable<\/p>\n
Yes \u2014 but licensed, not open-weight. Pokee AI serves Isaac through an OpenAI-compatible developer API<\/a>, and licenses it for deployment inside a VPC, on-premises, or on-device. The launch announcement advertises Day-0 support for vLLM and SGLang, and single-GPU serving starting from an RTX 4090 or equivalent. The research team publishes measurements only from a single B200-class GPU, so treat the consumer-GPU claim as vendor guidance rather than a reported result.<\/p>\n Company level: This fits organizations that already own their inference stack \u2014 mid-size and enterprise teams with a platform group, plus device OEMs. A solo practitioner without on-prem hardware should use the hosted API instead; the boundary argument only pays off if you have a boundary.<\/p>\n Industries: Healthcare and payors, financial services and insurance, defense and public sector, legal and e-discovery, and pharma or semiconductor R&D. The common trait is a rule that the data cannot cross an external API boundary, not a preference for privacy.<\/p>\n Applications: Whole-repository code review, multi-year contract and claims analysis, incident forensics over full log archives, and long-running tool agents that never need summarization or context pruning. The research paper makes this second point explicitly: when enough usable context is available in-boundary, memory hierarchies and compression become optional rather than required.<\/p>\n ‘+names[i]+’<\/p>\n ‘+ $’+costs[i].toFixed(0)+’<\/p>\n ‘; lc();sp();ag();dt();cm(); ” style=”width:100%;border:0;overflow:hidden;display:block” height=”600″ scrolling=”no” loading=”lazy”><\/p>\n Long-context results<\/p>\n On RULER<\/a>, Isaac stays above 93.3% at every tested length, ending at 93.3% at 10M. GPT-5.6 Luna and Gemini 3.5 Flash Lite track it to 512K, then hit context-overflow at 1M. <\/p>\n On MRCR v2<\/a> with 8 needles, Isaac scores 0.607, 0.743, and 0.500 at 256K, 512K, and 1M. Its margin over Gemini widens from 0.133 to 0.295 across that sweep. <\/p>\n Agentic and security results<\/p>\n Isaac leads BFCL v4<\/a> at 70.94 against Luna\u2019s 70.61. The report calls that parity rather than a lead, which is the correct read. On \u03c4\u00b3-bench<\/a> it averages 0.662 across four domains, ahead of Gemini\u2019s 0.631, with banking at 0.186 for everyone\u2019s difficulty. On MCP-Atlas<\/a> it places third at 74.59% coverage, but uses 9.10 turns per task against Gemini\u2019s 14.99. On Terminal-Bench 2.1<\/a> it resolves 56 of 86 text-compatible tasks (65.1%), behind Luna\u2019s 60. That is the one benchmark a cloud baseline wins, and the report states it plainly.<\/p>\n On DTAP<\/a> red-teaming, Isaac records the lowest direct (36.0), indirect (35.2), and combined (35.6) attack success rates, with 82.5 benign success. One condition differs: baselines ran under the stock runner, Isaac under the Pokee harness.<\/p>\n Efficiency, pricing, and portability<\/p>\n Under the RULER workload on one B200-class GPU, TTFT is 23.6s at 1M and 72.9s at 10M. Prefill throughput rises with context, from 42,400 to 137,200 tokens\/s, so a ten-fold longer prompt costs roughly three times the TTFT. List pricing is $0.15\/$1.00 per million input\/output tokens, marked provisional. Isaac also runs fully on-device on Intel Arc Pro B70 and Core Ultra Series 3 (Panther Lake), and on Qualcomm Snapdragon X2 Elite. <\/p>\n Key Takeaways<\/p>\n Pokee-Isaac 28B scores 93.3% on RULER at 10M tokens; every baseline in its panel returns 0.0 beyond 2M.<\/p>\n Prefill reaches 137,200 tokens\/s at 10M context on one B200; decode holds flat near 335 tokens\/s.<\/p>\n It leads BFCL v4 (70.94) and \u03c4\u00b3-bench (0.662 avg), places second on Terminal-Bench 2.1, third on MCP-Atlas.<\/p>\n Lowest combined attack success rate on DTAP (35.6) while keeping 82.5 benign task success.<\/p>\n Weights are not published; deployment is licensed into VPC, on-premises, or on-device.<\/p>\n Check out the\u00a0Blog<\/a> and Paper<\/a>.\u00a0Also,\u00a0feel free to follow us on\u00a0Twitter<\/a>\u00a0and don\u2019t forget to join our\u00a0150k+ML SubReddit<\/a>\u00a0and Subscribe to\u00a0our Newsletter<\/a>. Wait! are you on telegram?\u00a0now you can join us on telegram as well.<\/a><\/p>\n Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.?\u00a0Connect with us<\/a><\/p>\n
\n ”+
\n ‘<\/p>\n
\n }
\n $(‘#cmBars’).innerHTML=h;
\n requestAnimationFrame(function(){requestAnimationFrame(function(){
\n [].forEach.call($(‘#cmBars’).querySelectorAll(‘.fil’),function(f){f.style.width=f.getAttribute(‘data-w’)+’%’});
\n })});
\n var sav=costs[1]-costs[0];
\n $(‘#cmWhy’).innerHTML=’At ‘+vin+’M input tokens a month with ‘+rat+’% output, Isaac lists at $’+costs[0].toFixed(0)+
\n ‘ against $’+costs[1].toFixed(0)+’ for GPT-5.6 Luna and $’+costs[2].toFixed(0)+’ for Gemini 3.5 Flash Lite, a difference of $’+sav.toFixed(0)+
\n ‘ a month versus Luna. The report argues the per-token comparison still understates the gap, because Isaac is licensed to run inside the customer boundary, where cost is a fixed function of provisioned hardware rather than a variable function of tokens consumed.’;
\n}
\n$(‘#cmSlide’).oninput=cm;$(‘#cmRatio’).oninput=cm;<\/p>\n
\nfunction rz(){var h=document.getElementById(‘pk’).offsetHeight+40;
\n try{parent.postMessage({pkHeight:h},’*’)}catch(e){}}
\nsetInterval(rz,400);window.addEventListener(‘load’,rz);rz();
\n})();<\/p>\n
![\"\"](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/08\/a-professional-linkedin-headshot-photogr_0jcmb0R9Sv6nW5XK-zkPHw_uARV5VW1ST6osLNlunoVWg-150x150.png\")
<\/a><\/p>\n