OpenAI and Tencent announced parallel milestones this week in infrastructure governance and the opening of advanced models. On November 5, OpenAI published its National Community Pact and a Policy Blueprint for Compute Campuses, committing to fund its own energy, protect local resources, and create jobs. On November 4, Tencent launched Hy4, an open multimodal model with computer-use capabilities and integrated cybersecurity safeguards.
OpenAI: Community Pact and Compute Campus Blueprint
The National Community Pact establishes that each campus will operate under a local agreement with three pillars: pay for its own energy infrastructure, protect local resources and report on their impact, and generate opportunities for nearby workers and businesses. It requires financial guarantees before large investments, closed-loop water cooling plans, and publication of electrical and water consumption at the site level.
As concrete examples, in Effingham County (Georgia) they committed $4.8 million for free school meals for 14,000 students, and in Pike County (Ohio) $6 million together with SB Energy for daycare centers. The complementary Blueprint urges state commissions to require large consumers to cover the grid costs they add, and Congress to expedite transmission upgrades and tax incentives for clean electricity.
From an engineering perspective, the approach internalizes externalities—energy, water, noise, employment—into binding agreements and public audits, setting a precedent for other hyperscalers. The Blueprint is a policy proposal, not law; its adoption depends on regulators. Infrastructure teams should evaluate it as a framework when negotiating colocation or PPAs, demanding «grid self-pay» clauses and water and noise reporting. Source: OpenAI
Tencent Hy4: Open Model with Vision and Computer Use
Tencent positions Hy4 among the world’s leading open-source models, introducing visual understanding—image processing, chart analysis, and visual mathematical reasoning—and computer-use capabilities to interact with CLI and GUI, executing tasks in Blender, CAD, Unreal Engine, Unity, and Godot. It improves instruction following, long context, and long-horizon agentic tasks.
Internal benchmarks report an average score of 83.18 and a 96.15% success rate on WorkBuddy (12 scenarios, 9 top rankings in the Chinese market). In financial research (iRaB) it improves by 12.03 percentage points versus Preview. Hallucination drops from 12.95% to 7.2%, and errors from unsupported assumptions fall from 23.1% to 5.6%, indicating a reliability leap relevant for workflows demanding precision.
Availability spans Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom Dayuan, ima, and API via OpenRouter, Hermes, and OpenCode. Base pricing: 0.90 S$/M input tokens, 2.70 US$/M output tokens, 0.045 US$/M cache tokens; Hy4 Flash coming soon. Promotions: WorkBuddy and CodeBuddy free from November 4 to 15, TokenHub 50% off until Dec 31, 2026. Source: Tencent
Security by Design: Exclusion of Offensive Material and Audits
Hy4 was trained on defensive data—proof-of-concept exploits, remediation, reverse engineering, and pentesting for analysis—deliberately excluding offensive exploitation material. The internal Xuanwu Lab and Keen Security Lab evaluated the model and concluded that «no significant concerns about offensive capabilities were found in the areas evaluated,» establishing a «defensive-security-first» standard.
In biosecurity, Hy4 shows a solid base of biological knowledge, though open-ended experimental design remains more limited than in frontier models. For CISOs and security teams, the recommendation is to test Hy4 in a sandbox for reverse engineering, binary analysis, and defensive PoC development, verifying that the exclusion of offensive material persists in fine‑tunes. The labs only evaluated specific areas; they do not certify a complete absence of risk.
This strategy contrasts with the common practice of releasing models without auditing offensive capabilities. Documenting the supply chain—defensive data, Xuanwu/Keen audits—enables evidence of «security by design» for ISO 42001 or the EU AI Act, even though Hy4 lacks formal certification; documentation is the adopter’s responsibility. Buyers should take advantage of the TokenHub 50% promotion for low-cost pilots and compare cost per token against closed alternatives on real workloads.
Implications for Infrastructure and AI Governance
The convergence reveals a dual trend: major players assume direct responsibility for physical externalities (OpenAI) and demonstrate that opening advanced models does not require giving up security safeguards (Tencent). The Pact obligates internalizing grid, water, and energy costs, while Hy4 integrates vision, computer use, and long reasoning without known offensive capabilities.
For infrastructure governance, the Blueprint provides a negotiable framework: require ‘grid self-pay’ clauses, water and noise consumption reporting, and closed-loop cooling plans. For model governance, Hy4 raises the bar for competitors (Llama, DeepSeek, Qwen) by combining real agentic capabilities with offensive security auditing, pressuring regulators to define enforceable baselines.
On cost and accessibility, flat pricing despite performance leaps and aggressive promotions accelerate enterprise adoption in Asia and, via third-party APIs, globally. FinOps teams must compare cost/token against GPT‑4o or Claude 3.5 on real workloads, confirming standard pricing post-promotion. The future release of weights and system cards will enable independent audits validating performance and safety.
Actionable recommendations for technical teams
1. Infrastructure/Cloud: evaluate the Compute Campus Policy Blueprint as a reference when negotiating colocation or PPAs; require ‘grid self-pay’ clauses and water/noise reporting. Limit: the Blueprint is a proposal, not law; its adoption depends on regulators.
2. CISOs/Security: test Hy4 in sandbox for reverse engineering, binary analysis, and defensive PoC; verify persistence of offensive material exclusion in fine‑tunes. Condition: Xuanwu/Keen labs only evaluated specific areas.
3. Agent developers: leverage Hy4’s computer use (CLI/GUI) to automate legacy workflows (CI/CD, code migration, testing in game engines) without exposing proprietary API keys. Limit: GameDev/GameLogic benchmarks are internal; validate in own pipelines before production. 4. Procurement/FinOps: capture TokenHub 50% promotion (through 31‑Dec‑2026) for pilots; compare cost/token against closed alternatives. Condition: promotional pricing; confirm standard pricing post‑promo. 5. Compliance/Regulatory: document supply chain (defensive data, Xuanwu/Keen audits) to evidence ‘security by design’ for ISO 42001 or EU AI Act. Limit: Hy4 lacks formal certification; documentation is the adopter’s responsibility.
Structured reference for readers and AI agents
{
"format": "rafael-fuentes-article/1.0",
"language": "en",
"title": "OpenAI Publishes Its Community Pact and Tencent Launches Hy4 with Computer Use",
"summary": "OpenAI sets local commitments on energy, water, and employment; Tencent unveils Hy4, an open multimodal model with vision, computer use, and defensive security.",
"author": "Rafael Fuentes",
"publicationDate": "2026-11-06T06:01:35.088+01:00",
"eventDate": "2026-11-05",
"url": "https://falifuentes.com/rafael-fuentes-2026-11-06-am-en/",
"translation": {
"language": "es",
"url": "https://falifuentes.com/rafael-fuentes-2026-11-06-am-es/"
},
"topics": [
"AI",
"Artificial Intelligence",
"Compliance and Regulations",
"Cybersecurity",
"Cybersecurity News",
"DeepSeek",
"English",
"OpenAI",
"Supply Chain"
],
"sources": [
"https://openai.com/index/setting-new-standard-responsible-ai-infrastructure",
"https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models"
]
}