[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-thinking-machines-inkling-how-a-new-open-weight-ai-challenges-the-status-quo-en":3,"ArticleBody_SeoAv1Z5CnCKccWZ6k6CQyZJUA4qMLHgfe0w3iaovcw":221},{"article":4,"relatedArticles":192,"locale":66},{"id":5,"title":6,"slug":7,"content":8,"htmlContent":9,"excerpt":10,"category":11,"tags":12,"metaDescription":10,"wordCount":13,"readingTime":14,"publishedAt":15,"sources":16,"sourceCoverage":58,"transparency":60,"seo":63,"language":66,"featuredImage":67,"featuredImageCredit":68,"isFreeGeneration":72,"trendSlug":73,"trendSnapshot":74,"niche":82,"geoTakeaways":86,"geoFaq":95,"entities":105},"6a58313a5a245dc50f2b6e4e","Thinking Machines’ Inkling: How a New Open-Weight AI Challenges the Status Quo","thinking-machines-inkling-how-a-new-open-weight-ai-challenges-the-status-quo","[Inkling](\u002Fentities\u002F6a583275b15b2ddcc32c78f3-inkling) is [Thinking Machines](\u002Fentities\u002F69d5c7914eea09eba3e0d531-thinking-machines)’ first general-purpose open-weight model, launched as Western developers seek options beyond closed APIs and Chinese open-weight giants.[1][2] For technical leaders, the real issue is who controls the infrastructure [your agents and copilots](\u002Farticle\u002Fnvidia-s-nemoclaw-how-an-open-ai-agent-toolkit-will-reshape-enterprise-workflows) rely on, not which model tops a leaderboard.[3]  \n\n## 1. Inkling in Context: Why This Open-Weight Release Matters  \n\nInkling is an open-weight system:  \n\n- Full weights are downloadable  \n- Models can run on your own infrastructure  \n- Deep customization is possible instead of metered API dependence[1][2]  \n\nMost frontier assistants are still pay-per-token black boxes, so this alone is notable.  \n\nKey characteristics:  \n\n- Trained from scratch for native text, image, audio, and video use—not a text model with vision attached later[1][3]  \n- Positioned as a broadly capable workhorse, with strong reasoning and coding, rather than a benchmark trophy[1][3]  \n\n📊 **Data point**  \nInkling has 975 billion parameters, placing it among the largest Mixture-of-Experts (MoE) models publicly available.[2][3]  \n\nStrategic & geopolitical context:  \n\n- Today’s strongest open-weight models largely come from Chinese labs[2]  \n- [Meta](\u002Fentities\u002F6939b254312dc892c4c18581-meta)’s shift away from an open Llama 4 left a Western gap[2]  \n- Many cost-sensitive firms have standardized on Chinese systems as their primary alternative to expensive proprietary stacks[2]  \n\nThinking Machines instead argues for decentralized AI:  \n\n- No small group of vendors should control powerful models  \n- Models should be adaptable to local data, regulation, and governance regimes[1][3]  \n\n💡 **Key takeaway**  \nInkling matters less as a single model card and more as the return of a Western open-weight option near Chinese frontier levels, aligned with [US](\u002Fentities\u002F695de00619d266277e14dbd8-us) and EU governance expectations.[1][2][5]  \n\n## 2. Inside Inkling: Architecture, Scale, and Capabilities  \n\nCore architecture:  \n\n- MoE transformer: 975B total parameters, ~41B active per token[3]  \n- Each token uses a small expert subset, allowing:  \n  - Tunable “thinking effort”  \n  - Explicit quality vs. cost tradeoffs per request[3]  \n\nTraining and context:  \n\n- Pretrained on 45T multimodal tokens: text, images, audio, video[3]  \n- Native cross-format reasoning without external adapters[3]  \n- Up to 1M-token context window for:  \n  - Whole-repo code analysis  \n  - Large planning documents  \n  - Joint multimedia transcripts in one session[3]  \n\n⚠️ **Key point**  \nA 1M-token window increases risk: teams may pour entire systems into prompts, widening the blast radius of misrouting, leakage, or prompt injection if guardrails are weak.[3][10]  \n\nPerformance and positioning:  \n\n- [OpenAI](\u002Fentities\u002F6939892d312dc892c4c1841a-openai), [Anthropic](\u002Fentities\u002F6939b254312dc892c4c1857e-anthropic), and [Google](\u002Fentities\u002F6939b254312dc892c4c18580-google) closed models still lead modestly on many benchmarks[2][3]  \n- Inkling is competitive across domains and strong on agent-style tasks[2][3]  \n- Design goal: best open-weight base for domain adaptation, not “best model on earth”[1][3]  \n\nModel family:  \n\n- Inkling‑Small: ~12B active parameters with similar recipe[3]  \n- Targets lower latency and cost while keeping multimodal support[3]  \n\n💡 **Key takeaway**  \nInkling trades some peak benchmark rankings for controllability—over context, cost, and customization—making it attractive as a foundation model rather than a single turnkey product.[2][3]  \n\n## 3. Enterprise and Ecosystem Impact of an Open-Weight Giant  \n\nInkling integrates into [Tinker](\u002Fentities\u002F698950af033ff25c8c61c303-tinker), Thinking Machines’ customization platform, so teams can:  \n\n- Fine-tune variants  \n- Evaluate and deploy in a single loop[2][3]  \n\nTinker has already powered [Bridgewater Associates](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBridgewater_Associates)’ customized Qwen variant that beat top proprietary models on internal tasks at lower cost—evidence that major institutions will bet on tuned open-weights when economics favor them.[2]  \n\n📊 **Data point**  \nOver 55% of North American enterprises have deployed or are testing LLM tools, yet many face privacy, compliance, and reliability issues.[4] Open-weight models that can be privately hosted and deeply audited directly address these concerns.[1][3][4]  \n\nGovernance and control benefits:  \n\n- Sovereign deployments ease legal and compliance worries (“Where exactly is this data going?”)  \n- Full control of logs, red-teaming, and agent permissions for sensitive systems (payments, HR, trading)[3][10]  \n\n💼 **Enterprise angle**  \nOpen weights enable:  \n\n- Multi-model resilience to avoid single-vendor lock-in  \n- Private or jurisdiction-specific deployments for regulated workloads  \n- Richer, auditable agent sandboxes and human-in-the-loop approval flows[3][5][10]  \n\nRegulatory alignment:  \n\n- US states ([California](\u002Fentities\u002F69398a91312dc892c4c18449-california), [New York](\u002Fentities\u002F6939aeb8312dc892c4c1850a-new-york), Illinois) are converging on a de facto national regime for transparency, incident reporting, and audits of frontier systems.[5]  \n- Inspectable, tunable weights fit more naturally into these obligations than sealed cloud APIs.[1][3][5]  \n\n⚡ **Key point**  \nInkling is as much a governance building block as a technical artifact: it gives regulators, auditors, and enterprises something they can inspect, benchmark, and constrain.[3][5]  \n\n## Conclusion: Inkling’s Bet on Who Controls Frontier-Grade AI  \n\nInkling is not aimed at dethroning every closed frontier model on raw scores; it is aimed at shifting control.[1][2][3] With a 975B-parameter, multimodal, open-weight family built for customization, private hosting, and governed multi-model stacks, Thinking Machines challenges a status quo dominated by sealed APIs and Chinese open offerings.[1][2][3][10]  \n\nFor CTOs, CISOs, and founders, the next move is empirical: deploy Inkling via Tinker, compare it against your stack, and measure where open-weights improve cost, control, and compliance.[2][3][4] Treat it not as a single magic model, but as a backbone for secure, multi-model AI infrastructure you can own, audit, and evolve.","\u003Cp>\u003Ca href=\"\u002Fentities\u002F6a583275b15b2ddcc32c78f3-inkling\">Inkling\u003C\u002Fa> is \u003Ca href=\"\u002Fentities\u002F69d5c7914eea09eba3e0d531-thinking-machines\">Thinking Machines\u003C\u002Fa>’ first general-purpose open-weight model, launched as Western developers seek options beyond closed APIs and Chinese open-weight giants.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa> For technical leaders, the real issue is who controls the infrastructure \u003Ca href=\"\u002Farticle\u002Fnvidia-s-nemoclaw-how-an-open-ai-agent-toolkit-will-reshape-enterprise-workflows\" class=\"internal-link\">your agents and copilots\u003C\u002Fa> rely on, not which model tops a leaderboard.\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>1. Inkling in Context: Why This Open-Weight Release Matters\u003C\u002Fh2>\n\u003Cp>Inkling is an open-weight system:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Full weights are downloadable\u003C\u002Fli>\n\u003Cli>Models can run on your own infrastructure\u003C\u002Fli>\n\u003Cli>Deep customization is possible instead of metered API dependence\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Most frontier assistants are still pay-per-token black boxes, so this alone is notable.\u003C\u002Fp>\n\u003Cp>Key characteristics:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Trained from scratch for native text, image, audio, and video use—not a text model with vision attached later\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Positioned as a broadly capable workhorse, with strong reasoning and coding, rather than a benchmark trophy\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>📊 \u003Cstrong>Data point\u003C\u002Fstrong>\u003Cbr>\nInkling has 975 billion parameters, placing it among the largest Mixture-of-Experts (MoE) models publicly available.\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Strategic & geopolitical context:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Today’s strongest open-weight models largely come from Chinese labs\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"\u002Fentities\u002F6939b254312dc892c4c18581-meta\">Meta\u003C\u002Fa>’s shift away from an open Llama 4 left a Western gap\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Many cost-sensitive firms have standardized on Chinese systems as their primary alternative to expensive proprietary stacks\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Thinking Machines instead argues for decentralized AI:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>No small group of vendors should control powerful models\u003C\u002Fli>\n\u003Cli>Models should be adaptable to local data, regulation, and governance regimes\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>💡 \u003Cstrong>Key takeaway\u003C\u002Fstrong>\u003Cbr>\nInkling matters less as a single model card and more as the return of a Western open-weight option near Chinese frontier levels, aligned with \u003Ca href=\"\u002Fentities\u002F695de00619d266277e14dbd8-us\">US\u003C\u002Fa> and EU governance expectations.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>2. Inside Inkling: Architecture, Scale, and Capabilities\u003C\u002Fh2>\n\u003Cp>Core architecture:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>MoE transformer: 975B total parameters, ~41B active per token\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Each token uses a small expert subset, allowing:\n\u003Cul>\n\u003Cli>Tunable “thinking effort”\u003C\u002Fli>\n\u003Cli>Explicit quality vs. cost tradeoffs per request\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Training and context:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Pretrained on 45T multimodal tokens: text, images, audio, video\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Native cross-format reasoning without external adapters\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Up to 1M-token context window for:\n\u003Cul>\n\u003Cli>Whole-repo code analysis\u003C\u002Fli>\n\u003Cli>Large planning documents\u003C\u002Fli>\n\u003Cli>Joint multimedia transcripts in one session\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>⚠️ \u003Cstrong>Key point\u003C\u002Fstrong>\u003Cbr>\nA 1M-token window increases risk: teams may pour entire systems into prompts, widening the blast radius of misrouting, leakage, or prompt injection if guardrails are weak.\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-10\" class=\"citation-link\" title=\"View source [10]\">[10]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Performance and positioning:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Ca href=\"\u002Fentities\u002F6939892d312dc892c4c1841a-openai\">OpenAI\u003C\u002Fa>, \u003Ca href=\"\u002Fentities\u002F6939b254312dc892c4c1857e-anthropic\">Anthropic\u003C\u002Fa>, and \u003Ca href=\"\u002Fentities\u002F6939b254312dc892c4c18580-google\">Google\u003C\u002Fa> closed models still lead modestly on many benchmarks\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Inkling is competitive across domains and strong on agent-style tasks\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Design goal: best open-weight base for domain adaptation, not “best model on earth”\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Model family:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Inkling‑Small: ~12B active parameters with similar recipe\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Targets lower latency and cost while keeping multimodal support\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>💡 \u003Cstrong>Key takeaway\u003C\u002Fstrong>\u003Cbr>\nInkling trades some peak benchmark rankings for controllability—over context, cost, and customization—making it attractive as a foundation model rather than a single turnkey product.\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>3. Enterprise and Ecosystem Impact of an Open-Weight Giant\u003C\u002Fh2>\n\u003Cp>Inkling integrates into \u003Ca href=\"\u002Fentities\u002F698950af033ff25c8c61c303-tinker\">Tinker\u003C\u002Fa>, Thinking Machines’ customization platform, so teams can:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Fine-tune variants\u003C\u002Fli>\n\u003Cli>Evaluate and deploy in a single loop\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Tinker has already powered \u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBridgewater_Associates\" class=\"wiki-link\" target=\"_blank\" rel=\"noopener\">Bridgewater Associates\u003C\u002Fa>’ customized Qwen variant that beat top proprietary models on internal tasks at lower cost—evidence that major institutions will bet on tuned open-weights when economics favor them.\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>📊 \u003Cstrong>Data point\u003C\u002Fstrong>\u003Cbr>\nOver 55% of North American enterprises have deployed or are testing LLM tools, yet many face privacy, compliance, and reliability issues.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> Open-weight models that can be privately hosted and deeply audited directly address these concerns.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Governance and control benefits:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Sovereign deployments ease legal and compliance worries (“Where exactly is this data going?”)\u003C\u002Fli>\n\u003Cli>Full control of logs, red-teaming, and agent permissions for sensitive systems (payments, HR, trading)\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-10\" class=\"citation-link\" title=\"View source [10]\">[10]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>💼 \u003Cstrong>Enterprise angle\u003C\u002Fstrong>\u003Cbr>\nOpen weights enable:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Multi-model resilience to avoid single-vendor lock-in\u003C\u002Fli>\n\u003Cli>Private or jurisdiction-specific deployments for regulated workloads\u003C\u002Fli>\n\u003Cli>Richer, auditable agent sandboxes and human-in-the-loop approval flows\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003Ca href=\"#source-10\" class=\"citation-link\" title=\"View source [10]\">[10]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Regulatory alignment:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>US states (\u003Ca href=\"\u002Fentities\u002F69398a91312dc892c4c18449-california\">California\u003C\u002Fa>, \u003Ca href=\"\u002Fentities\u002F6939aeb8312dc892c4c1850a-new-york\">New York\u003C\u002Fa>, Illinois) are converging on a de facto national regime for transparency, incident reporting, and audits of frontier systems.\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Inspectable, tunable weights fit more naturally into these obligations than sealed cloud APIs.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>⚡ \u003Cstrong>Key point\u003C\u002Fstrong>\u003Cbr>\nInkling is as much a governance building block as a technical artifact: it gives regulators, auditors, and enterprises something they can inspect, benchmark, and constrain.\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>Conclusion: Inkling’s Bet on Who Controls Frontier-Grade AI\u003C\u002Fh2>\n\u003Cp>Inkling is not aimed at dethroning every closed frontier model on raw scores; it is aimed at shifting control.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa> With a 975B-parameter, multimodal, open-weight family built for customization, private hosting, and governed multi-model stacks, Thinking Machines challenges a status quo dominated by sealed APIs and Chinese open offerings.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-10\" class=\"citation-link\" title=\"View source [10]\">[10]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>For CTOs, CISOs, and founders, the next move is empirical: deploy Inkling via Tinker, compare it against your stack, and measure where open-weights improve cost, control, and compliance.\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> Treat it not as a single magic model, but as a backbone for secure, multi-model AI infrastructure you can own, audit, and evolve.\u003C\u002Fp>\n","Inkling is Thinking Machines’ first general-purpose open-weight model, launched as Western developers seek options beyond closed APIs and Chinese open-weight giants.[1][2] For technical leaders, the r...","trend-radar",[],797,4,"2026-07-16T01:26:01.779Z",[17,22,26,30,34,38,42,46,50,54],{"title":18,"url":19,"summary":20,"type":21},"Thinking Machines Lab Drops Its First Model Inkling","https:\u002F\u002Fwww.wired.com\u002Fstory\u002Fthinking-machines-lab-releases-its-first-model-inkling\u002F","Thinking Machines Lab, an artificial intelligence company started by exiles from OpenAI, has released its first model, called Inkling. The startup’s new model is open-weight, which means that research...","kb",{"title":23,"url":24,"summary":25,"type":21},"AI startup Thinking Machines launches an open-weight AI model","https:\u002F\u002Fwww.reuters.com\u002Fbusiness\u002Fai-startup-thinking-machines-launches-an-open-weight-ai-model-2026-07-15\u002F","SAN FRANCISCO, July 15 (Reuters) - AI startup Thinking Machines revealed on Wednesday a new artificial intelligence model that could serve as one of the few alternatives to popular open-source offerin...",{"title":27,"url":28,"summary":29,"type":21},"Inkling: Our open-weights model","https:\u002F\u002Fthinkingmachines.ai\u002Fnews\u002Fintroducing-inkling\u002F","Jul 15, 2026\n\nOur mission is to build AI that extends human will and judgment. We have developed a platform that lets anyone customize models, previewed an AI system built for interactive collaboratio...",{"title":31,"url":32,"summary":33,"type":21},"Enterprise LLM Integration: Challenges, Solutions, and Best Practices","https:\u002F\u002Fwww.nitorinfotech.com\u002Fblog\u002Fenterprise-llm-integration-challenges-and-best-practices\u002F","Enterprise LLM Integration: Challenges, Solutions, and Best Practices\n\nAbout the author\nNitor User\nArtificial intelligence | 22 Oct 2025 | 20 min\n\nHighlights\nIntegrating large language models (LLMs) i...",{"title":35,"url":36,"summary":37,"type":21},"Illinois Imposes Transparency and Safety Obligations on Frontier AI Systems","https:\u002F\u002Fwww.crowell.com\u002Fen\u002Finsights\u002Fclient-alerts\u002Fillinois-imposes-transparency-and-safety-obligations-on-frontier-ai-systems","Illinois Imposes Transparency and Safety Obligations on Frontier AI Systems\n\nWhat You Need to Know\n\n- Key takeaway #1\nIllinois has joined California and New York in adopting legislation to establish t...",{"title":39,"url":40,"summary":41,"type":21},"OpenAI’s first hardware device is reportedly a screenless speaker that can move","https:\u002F\u002Ftechcrunch.com\u002F2026\u002F07\u002F14\u002Fopenais-first-hardware-device-is-reportedly-a-screenless-speaker-that-can-move\u002F","Lucas Ropek • 3:22 PM PDT · July 14, 2026\n\nOpenAI’s first foray into hardware devices is reported to be a mobile smart speaker with integrated AI capabilities that can sync with ChatGPT and provide ot...",{"title":43,"url":44,"summary":45,"type":21},"OpenAI Targets 2026 AI Speaker Launch as Apple Lawsuit Looms","https:\u002F\u002Ffinance.yahoo.com\u002Ftechnology\u002Fai\u002Farticles\u002Fopenai-targets-2026-ai-speaker-151433211.html","OpenAI, a leading developer of artificial intelligence models, is preparing to enter the consumer-device market with a mobile, screen-free smart speaker designed to bring ChatGPT into the home. The pr...",{"title":47,"url":48,"summary":49,"type":21},"Everybody's AI Project to develop domestic AI services and launch free nationwide AI chatbot","https:\u002F\u002Fwww.mk.co.kr\u002Fen\u002Fit\u002F12097161","The government will release a domestic artificial intelligence (AI) chatbot within this year that anyone can use without paying and without any restrictions on the amount of use. It aims to make inter...",{"title":51,"url":52,"summary":53,"type":21},"OpenAI's First Device Will Be Moveable Screenless Speaker Built as AI Companion","https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-07-14\u002Fopenai-s-first-device-will-be-moveable-screenless-speaker-built-as-ai-companion","By Mark Gurman\n\nJuly 14, 2026 at 8:40 PM UTC\n\nTakeaways by Bloomberg AI: OpenAI’s much-anticipated push into consumer devices is slated to begin with a mobile, screen-free smart speaker designed to be...",{"title":55,"url":56,"summary":57,"type":21},"Building secure and governed AI systems beyond prompt engineering","https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002F538424580633983\u002Fposts\u002F1694067088403054\u002F","Building secure and governed AI systems beyond prompt engineering.\n\nThe AI shift is no longer about writing better prompts. That was phase one. The next phase is about secure orchestration.\n\nTeams now...",{"totalSources":59},10,{"generationDuration":61,"kbQueriesCount":59,"confidenceScore":62,"sourcesCount":59},268530,100,{"metaTitle":64,"metaDescription":65},"Inkling Open-Weight AI: Decentralizing Model Control","Rethink control: Inkling is a downloadable open-weight AI for private infrastructure, enabling customization and a Western alternative—learn what changes","en","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1618939291225-8d558ea4369f?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHx0aGlua2luZyUyMG1hY2hpbmVzJTIwb3BlbiUyMHdlaWdodHxlbnwxfDB8fHwxNzg0MTY0NjY2fDA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60",{"photographerName":69,"photographerUrl":70,"unsplashUrl":71},"Quilia","https:\u002F\u002Funsplash.com\u002F@heyquilia?utm_source=coreprose&utm_medium=referral","https:\u002F\u002Funsplash.com\u002Fphotos\u002Fblack-and-silver-electronic-device-5ddH9Y2accI?utm_source=coreprose&utm_medium=referral",true,"thinking-machines-open-weight-ai-model-release",{"score":62,"type":75,"sourceCount":76,"topSourceDomains":77,"detectedAt":81,"mentionsLast7Days":76},"spiking",11,[78,79,80],"unknown","reuters.com","wired.com","2026-07-16T00:38:06.252Z",{"key":83,"name":84,"nameEn":85},"ia","Intelligence Artificielle","Artificial Intelligence",[87,89,91,93],{"text":88},"Inkling is a 975 billion-parameter open-weight Mixture-of-Experts (MoE) model with downloadable weights and ~41B active parameters per token, enabling private hosting and deep customization.",{"text":90},"Inkling supports native multimodal training (text, image, audio, video) with a 1,000,000-token context window and was pretrained on 45 trillion multimodal tokens.",{"text":92},"Thinking Machines positions Inkling as a governance-first foundation model: it prioritizes controllability, private deployment, and auditability over narrow leaderboard dominance.",{"text":94},"Inkling integrates with the Tinker platform for in‑loop fine-tuning and deployment, enabling enterprises to reduce vendor lock-in and meet jurisdictional compliance needs.",[96,99,102],{"question":97,"answer":98},"What is the practical advantage of Inkling being open-weight versus closed API models?","Open-weight availability means you can download Inkling’s full weights and run the model on your own infrastructure, enabling full control over data flows, logs, and model updates. This reduces reliance on metered cloud APIs, lowers long-term costs for heavy usage, and allows organizations to perform deep customization, domain adaptation, and red‑teaming with direct access to model internals. For regulated or privacy-sensitive workloads, hosting the model privately simplifies compliance and incident auditing compared with sealed, remote models.",{"question":100,"answer":101},"How does Inkling’s 1M-token context window change enterprise usage and risk?","A 1,000,000-token context window lets teams analyze entire codebases, long legal or technical documents, and joint multimedia transcripts in a single session, improving agent planning, retrieval-augmented workflows, and end-to-end reasoning. However, it increases the blast radius for data leakage, misrouting, and prompt injection if guardrails are weak, so enterprises must pair Inkling with strict access controls, input\u002Foutput sanitization, and thorough red‑teaming to avoid unintended exposure of sensitive systems or PII.",{"question":103,"answer":104},"Where does Inkling sit on performance and who should adopt it?","Inkling is competitive across domains and particularly strong for agent-style tasks, but closed models from major cloud providers still modestly lead on some benchmarks; Inkling’s strategic advantage is controllability rather than raw top-line scores. Organizations that prioritize private hosting, auditability, multi-model resilience, or jurisdictional sovereignty—such as finance, healthcare, and regulated government contractors—should adopt Inkling as a foundation model for customization, cost optimization, and compliance-driven deployments.",[106,114,120,125,133,139,146,153,159,165,171,177,182,187],{"id":107,"name":108,"type":109,"confidence":110,"wikipediaUrl":111,"slug":112,"mentionCount":113},"6998c0e99aa9beba177c7874","open-weight model","concept",0.95,null,"6998c0e99aa9beba177c7874-open-weight-model",2,{"id":115,"name":116,"type":109,"confidence":117,"wikipediaUrl":111,"slug":118,"mentionCount":119},"6a583351b15b2ddcc32c7964","MoE transformer",0.94,"6a583351b15b2ddcc32c7964-moe-transformer",1,{"id":121,"name":122,"type":109,"confidence":123,"wikipediaUrl":111,"slug":124,"mentionCount":119},"6a583351b15b2ddcc32c7966","decentralized AI",0.88,"6a583351b15b2ddcc32c7966-decentralized-ai",{"id":126,"name":127,"type":128,"confidence":129,"wikipediaUrl":130,"slug":131,"mentionCount":132},"69398a91312dc892c4c18449","California","location",0.99,"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCalifornia","69398a91312dc892c4c18449-california",220,{"id":134,"name":135,"type":128,"confidence":129,"wikipediaUrl":136,"slug":137,"mentionCount":138},"6939aeb8312dc892c4c1850a","New York","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FNew_York_City","6939aeb8312dc892c4c1850a-new-york",64,{"id":140,"name":141,"type":128,"confidence":142,"wikipediaUrl":143,"slug":144,"mentionCount":145},"695de00619d266277e14dbd8","US",0.98,"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FUs","695de00619d266277e14dbd8-us",6,{"id":147,"name":148,"type":149,"confidence":129,"wikipediaUrl":150,"slug":151,"mentionCount":152},"6939892d312dc892c4c1841a","OpenAI","organization","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FOpenAI","6939892d312dc892c4c1841a-openai",873,{"id":154,"name":155,"type":149,"confidence":129,"wikipediaUrl":156,"slug":157,"mentionCount":158},"6939b254312dc892c4c1857e","Anthropic","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FAnthropic","6939b254312dc892c4c1857e-anthropic",545,{"id":160,"name":161,"type":149,"confidence":129,"wikipediaUrl":162,"slug":163,"mentionCount":164},"6939b254312dc892c4c18580","Google","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FGoogle","6939b254312dc892c4c18580-google",539,{"id":166,"name":167,"type":149,"confidence":129,"wikipediaUrl":168,"slug":169,"mentionCount":170},"6939b254312dc892c4c18581","Meta","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FMeta","6939b254312dc892c4c18581-meta",184,{"id":172,"name":173,"type":149,"confidence":174,"wikipediaUrl":175,"slug":176,"mentionCount":14},"698950af033ff25c8c61c303","Tinker",0.97,"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FTinker","698950af033ff25c8c61c303-tinker",{"id":178,"name":179,"type":149,"confidence":129,"wikipediaUrl":180,"slug":181,"mentionCount":14},"69d5c7914eea09eba3e0d531","Thinking Machines","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FThinking_Machines_Lab","69d5c7914eea09eba3e0d531-thinking-machines",{"id":183,"name":184,"type":149,"confidence":110,"wikipediaUrl":185,"slug":186,"mentionCount":113},"6a5832d7b15b2ddcc32c794b","Bridgewater Associates","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBridgewater_Associates","6a5832d7b15b2ddcc32c794b-bridgewater-associates",{"id":188,"name":189,"type":149,"confidence":190,"wikipediaUrl":111,"slug":191,"mentionCount":119},"6a583352b15b2ddcc32c7968","China-based labs",0.8,"6a583352b15b2ddcc32c7968-china-based-labs",[193,200,207,214],{"id":194,"title":195,"slug":196,"excerpt":197,"category":11,"featuredImage":198,"publishedAt":199},"6a5860ae5a245dc50f2b748a","Illinois Guidance on AI Use in Public Schools: What Districts Need to Know Now","illinois-guidance-on-ai-use-in-public-schools-what-districts-need-to-know-now","In 2025, only 26 states had issued K‑12 AI guidance, and just 18% of principals reported any direction from their school or district—dropping to 13% in high‑poverty schools. [6] For Illinois educators...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1673241564420-9ca6abde6a0b?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxpbGxpbm9pcyUyMGd1aWRhbmNlJTIwdXNlJTIwcHVibGljfGVufDF8MHx8fDE3ODQxNzY4MTR8MA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-16T04:46:51.723Z",{"id":201,"title":202,"slug":203,"excerpt":204,"category":11,"featuredImage":205,"publishedAt":206},"6a57c28b5a245dc50f2b520b","OpenAI’s Screen‑Free Mobile AI Speaker: What the 2026 Launch Means for the Future of Home Computing","openai-s-screen-free-mobile-ai-speaker-what-the-2026-launch-means-for-the-future-of-home-computing","For years, smart speakers have mostly worked as voice remotes. OpenAI’s planned mobile, screen‑free AI speaker aims to act more like a home computer for the conversational era, built around large lang...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1676299081847-824916de030a?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxvcGVuYWklMjBtb2JpbGUlMjBzY3JlZW4lMjBmcmVlfGVufDF8MHx8fDE3ODQxMzYzMzF8MA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-15T17:34:57.211Z",{"id":208,"title":209,"slug":210,"excerpt":211,"category":11,"featuredImage":212,"publishedAt":213},"6a5768d9b14fe5915b3ecf3e","Emerging Enterprise and Insurance Risks from Rapid AI Adoption","emerging-enterprise-and-insurance-risks-from-rapid-ai-adoption","1. How Rapid AI Adoption Is Rewiring Enterprise Risk\n\nEnterprise AI has moved into production: operational use jumped from 45% to 63% in a year, yet fewer than half of organizations have a formal AI r...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1637763723578-79a4ca9225f7?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxlbWVyZ2luZyUyMGVudGVycHJpc2UlMjBpbnN1cmFuY2UlMjByaXNrc3xlbnwxfDB8fHwxNzg0MTEzMzY5fDA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-15T11:11:10.916Z",{"id":215,"title":216,"slug":217,"excerpt":218,"category":11,"featuredImage":219,"publishedAt":220},"6a54d65c1d32dce929a1107b","Inside South Korea’s Plan for a Free Nationwide Government AI Chatbot","inside-south-korea-s-plan-for-a-free-nationwide-government-ai-chatbot","South Korea is about to run a national experiment: what happens when a government turns advanced conversational AI into a public utility, as basic as water or electricity?  \n\nUnder the “Everybody’s AI...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1619179834700-7a886aac80cc?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxzb3V0aCUyMGtvcmVhbiUyMGdvdmVybm1lbnQlMjByZWxlYXNlfGVufDF8MHx8fDE3ODM5NDQ3OTZ8MA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-13T12:23:09.515Z",["Island",222],{"key":223,"params":224,"result":226},"ArticleBody_SeoAv1Z5CnCKccWZ6k6CQyZJUA4qMLHgfe0w3iaovcw",{"props":225},"{\"articleId\":\"6a58313a5a245dc50f2b6e4e\"}",{"head":227},{}]