[{"@context":"https:\/\/schema.org\/","@type":"BlogPosting","@id":"https:\/\/blog.terabox.com\/insights\/uk-sovereign-ai-cosine-frontier-llm#BlogPosting","mainEntityOfPage":"https:\/\/blog.terabox.com\/insights\/uk-sovereign-ai-cosine-frontier-llm","headline":"UK Sovereign AI: How Cosine is Building a Frontier LLM","name":"UK Sovereign AI: How Cosine is Building a Frontier LLM","description":"\ud83d\udcfa Today’s recommended deep-dive video: https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk The Sovereign AI Race: How Britain Plans to Build Its Own Frontier LLMThe Rise of Sovereign IntelligenceArchitecture and the “Millions vs. Billions” ProblemSolving the “Spaghetti Monster” with RLKey TakeawaysQ&A The Sovereign AI Race: How... ","datePublished":"2026-07-23","dateModified":"2026-07-23","author":{"@type":"Person","@id":"https:\/\/blog.terabox.com\/author\/flextech-admin\/#Person","name":"flextech-admin","url":"https:\/\/blog.terabox.com\/author\/flextech-admin\/","image":{"@type":"ImageObject","@id":"https:\/\/secure.gravatar.com\/avatar\/ad516503a11cd5ca435acc9bb6523536?s=150&d=mm&r=gforcedefault=1","url":"https:\/\/secure.gravatar.com\/avatar\/ad516503a11cd5ca435acc9bb6523536?s=150&d=mm&r=gforcedefault=1","height":96,"width":96}},"publisher":{"@type":"Organization","name":"terabox","logo":{"@type":"ImageObject","@id":"http:\/\/blog.terabox.com\/wp-content\/uploads\/2021\/11\/logo\u4ea7\u54c1\u540d-\u7ad6\u7248.png","url":"http:\/\/blog.terabox.com\/wp-content\/uploads\/2021\/11\/logo\u4ea7\u54c1\u540d-\u7ad6\u7248.png","width":900,"height":900}},"image":{"@type":"ImageObject","@id":"https:\/\/img.youtube.com\/vi\/JTHmrELSfvk\/maxresdefault.jpg","url":"https:\/\/img.youtube.com\/vi\/JTHmrELSfvk\/maxresdefault.jpg","height":"","width":""},"url":"https:\/\/blog.terabox.com\/insights\/uk-sovereign-ai-cosine-frontier-llm","video":{"@context":"http:\/\/schema.org\/","@type":"VideoObject","@id":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk#VideoObject","contentUrl":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk","name":"Watching America Run Away With AI - Alistair Pullen (Cosine AI)","description":"This episode is sponsored by Notion. Learn more about Notion's Developer Platform today at https:\/\/notion.com\/mlst\n\nBritain's most capable coding model can't be exported, and that ban is the whole reason Cosine set out to build one from scratch. Alistair Pullen, CEO and co-founder of Cosine, sits down with Tim Scarfe to explain how a frontier system he calls Fable, locked behind US export controls, became the founding case for a UK sovereign model trained on the Isambard supercomputer in Bristol.\n\nThe bet underneath it is economic. Pullen argues that an inference company, rather than a training-first lab, doesn't need billions to compete: millions, a national compute allocation, and a consortium feedback loop can be enough. From there it gets into the machinery, why open-weight models still trail the frontier on size, active parameters and data, the mixture-of-experts versus dense trade-off and why active params dominate how a model actually feels, and the edge that real coding trajectories confer.\n\nThe back half is about making agents trustworthy. Pullen makes the case for beating \"slop\" by rewarding the process instead of the final answer, reframes code review as runtime proof (spin the bug up in a VM and force the agent to actually exploit it), and walks through Swarm, Cosine's system running hundreds of sub-agents in one shot. It ends on why memory is still an unsolved hack, how synthetic graders let you run RL on tasks with no built-in test, and why Pullen reads US export controls as an accidental gift, with a supply-chain sting in the tail.\n\n---\nTIMESTAMPS:\n00:00:00 The sovereign mandate and the Fable ban\n00:04:02 Millions vs billions: the inference-company model\n00:07:19 The consortium feedback loop\n00:07:40 Why open models lag the frontier\n00:14:59 MoE vs dense, and why active params matter\n00:16:29 Trajectories: the process-data advantage\n00:19:48 Beating slop: reward the process, not the answer\n00:26:06 Reusable abstractions and the epistemic wall\n00:29:56 Code review becomes runtime proof\n00:37:32 Do agentic harnesses still matter?\n00:40:35 Swarm: orchestrating hundreds of sub-agents\n00:45:14 Why memory is still unsolved\n00:48:25 Synthetic data and graders for RL\n00:53:09 The US export gift and supply-chain risk\n\n---\nREFERENCES:\norganization:\n[00:01:15] Cosine\nhttps:\/\/cosine.sh\n[00:04:14] Mistral AI\nhttps:\/\/mistral.ai\n[00:05:50] Anthropic\nhttps:\/\/www.anthropic.com\n[00:07:42] Cohere\nhttps:\/\/cohere.com\n[00:08:36] DeepSeek\nhttps:\/\/www.deepseek.com\ntool:\n[00:02:52] Isambard-AI\nhttps:\/\/isambard.ac.uk\n[00:05:56] Colossus (xAI)\nhttps:\/\/en.wikipedia.org\/wiki\/Colossus_(supercomputer)\n[00:07:52] GLM (Z.ai)\nhttps:\/\/z.ai\n[00:11:52] NVIDIA B300\nhttps:\/\/www.nvidia.com\/en-us\/data-center\/dgx-b300\/\n[00:15:37] gpt-oss-120b\nhttps:\/\/huggingface.co\/openai\/gpt-oss-120b\n[00:15:52] Devstral 2\nhttps:\/\/mistral.ai\/news\/devstral\n[00:16:01] Llama 70b\nhttps:\/\/www.llama.com\n[00:17:05] Claude Code\nhttps:\/\/www.anthropic.com\/claude-code\n[00:26:23] ARC-AGI (Francois Chollet)\nhttps:\/\/arcprize.org\n[00:40:38] Swarm (Cosine)\nhttps:\/\/cosine.sh\n[00:40:50] OpenAI Codex\nhttps:\/\/github.com\/openai\/codex\n[00:41:16] Lumen Outpost (Cosine)\nhttps:\/\/cosine.sh\n[00:41:18] Kimi K2 (Moonshot)\nhttps:\/\/huggingface.co\/moonshotai\/Kimi-K2-Instruct\n[00:49:55] SWE-bench\nhttps:\/\/www.swebench.com\n[00:52:40] SystemVerilog\nhttps:\/\/en.wikipedia.org\/wiki\/SystemVerilog\nperson:\n[00:23:40] Andrej Karpathy\nhttps:\/\/karpathy.ai\npaper:\n[00:27:10] GRPO (DeepSeekMath)\nhttps:\/\/arxiv.org\/abs\/2402.03300\n[00:27:13] GSPO\nhttps:\/\/arxiv.org\/abs\/2507.18071\n\nIncompressible Knowledge Probes, Bojie Li\nhttps:\/\/arxiv.org\/pdf\/2604.24827\n\nEstimating the Size of Claude Opus 4.5\/4.6\nhttps:\/\/unexcitedneurons.substack.com\/p\/estimating-the-size-of-claude-opus\n\n---\nReScript:\nhttps:\/\/app.rescript.info\/session\/5852d2b884c4ce4b?share=10b9799160845bb11779f8ac6cd3124f","thumbnailUrl":["https:\/\/i.ytimg.com\/vi\/JTHmrELSfvk\/default.jpg","https:\/\/i.ytimg.com\/vi\/JTHmrELSfvk\/mqdefault.jpg","https:\/\/i.ytimg.com\/vi\/JTHmrELSfvk\/hqdefault.jpg","https:\/\/i.ytimg.com\/vi\/JTHmrELSfvk\/sddefault.jpg","https:\/\/i.ytimg.com\/vi\/JTHmrELSfvk\/maxresdefault.jpg"],"uploadDate":"2026-07-13T22:23:38+00:00","duration":"PT55M57S","embedUrl":"https:\/\/www.youtube.com\/embed\/JTHmrELSfvk","publisher":{"@type":"Organization","@id":"https:\/\/www.youtube.com\/channel\/UCMLtBahI5DMrt0NPvDSoIRQ#Organization","url":"https:\/\/www.youtube.com\/channel\/UCMLtBahI5DMrt0NPvDSoIRQ","name":"Machine Learning Street Talk","description":"MLST is the leading highly technical AI podcast. Subscribe now! Welcome! We bring you the latest in advanced AI research, from the best AI experts in the world. Our approach is unrivalled in terms of scope and rigour \u2013 we believe in diversity of ideas (which is to say, not just LLMs!) and we also cover other promising alternative paths to AGI, as well as CogSci, CompSci, Neuro, Mathematics, Philosophy of Mind and Language.  \n\nSupport us on Patreon for early access, exclusive content, private Discord, biweekly calls and much more! \nhttps:\/\/www.patreon.com\/mlst \n\nDonate here: https:\/\/www.paypal.com\/donate\/?hosted_button_id=K2TYRVPBGXVNA\n\nPlease email us to learn about sponsorship packages and deals. \ntim at mlst.ai (please put your budget in the subject line)\n\nPodcast booking agencies - *don't contact us* - we wouldn't even interview anyone who needed a booking agent.\nMedia\/influence agencies - *don't contact us* - we only work directly with brands\/sponsors.\n","logo":{"url":"https:\/\/yt3.ggpht.com\/15Akj76BG8IsM5ctgqVwKXArl6IfIVFAbuGa1kOomoioRgJgXHHaLmMAW7iHTMRUoEfyjTtq8lg=s800-c-k-c0x00ffffff-no-rj","width":800,"height":800,"@type":"ImageObject","@id":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk#VideoObject_publisher_logo_ImageObject"}},"potentialAction":{"@type":"SeekToAction","@id":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk#VideoObject_potentialAction","target":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk&t={seek_to_second_number}","startOffset-input":"required name=seek_to_second_number"},"interactionStatistic":[[{"@type":"InteractionCounter","@id":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk#VideoObject_interactionStatistic_WatchAction","interactionType":{"@type":"WatchAction"},"userInteractionCount":6807}],{"@type":"InteractionCounter","@id":"https:\/\/www.youtube.com\/watch?v=JTHmrELSfvk#VideoObject_interactionStatistic_LikeAction","interactionType":{"@type":"LikeAction"},"userInteractionCount":148}]},"about":["\u300eEnglish\u300f","Insights"],"wordCount":1360,"keywords":["Creative writing","data security"]},{"@context":"https:\/\/schema.org\/","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Insights","item":"https:\/\/blog.terabox.com\/insights\/#breadcrumbitem"},{"@type":"ListItem","position":2,"name":"UK Sovereign AI: How Cosine is Building a Frontier LLM","item":"https:\/\/blog.terabox.com\/insights\/uk-sovereign-ai-cosine-frontier-llm#breadcrumbitem"}]}]