SunBrief#89:Chinese AI Model Kimi K3 Takes on Fable and GPT 5.6

Welcome to the SunBrief

Today in SunBrief 🌞

• 100+ ChatGPT Prompts to Revolutionize Your Day

• Chinese AI lab Moonshot AI released Kimi K3

• Stock Updates

• AI Brain Implant Helps Paralysed Man Regain Movement and Touch

• AWS Billing Bug Shows Customers Owe Billions in False Charges

• Get Your Personalized Blueprint for Health

• AI Highlights of the Week

• Too Important to Miss

100+ ChatGPT Prompts to Revolutionize Your Day

Stop wasting time figuring out what to ask ChatGPT. This free HubSpot guide gives you 100+ proven prompts to speed up writing, research, planning, and daily tasks.

From better outputs to faster workflows, this is a simple shortcut to getting more value from ChatGPT. Instant access. Real use cases. Prompts you’ll actually use.

Chinese AI lab Moonshot AI released Kimi K3

2.8-trillion-parameter multimodal model targets long-horizon coding, research, and agentic knowledge work

Moonshot AI has introduced Kimi K3, its most capable model yet, and the first open model in the 3-trillion-parameter class. It combines native vision, a 1-million-token context window, and a sparse mixture-of-experts architecture to deliver advanced agentic capabilities.

Key Points:

• Massive but sparse architecture: Kimi K3 has 2.8 trillion total parameters while activating only 16 of 896 experts, improving efficiency compared with Kimi K2.

• Long-horizon coding: The model can work autonomously across large repositories, optimize GPU kernels, build compilers, develop applications, and perform complex code migrations.

• Scientific and engineering work: K3 reportedly built a simulated chip for its own architecture and reproduced advanced astrophysics research that would normally take experts weeks.

• Native multimodal intelligence: It can reason across text, images, and video, supporting visual coding, 3D game creation, CAD-style workflows, and video editing.

• Agentic knowledge work: Kimi K3 can coordinate parallel subagents, conduct thousands of searches, analyze large document collections, and create interactive research reports, dashboards, and presentations.

• Open-weights release: The model is available through Kimi’s apps, Kimi Code, and API, with full weights scheduled for release by July 27, 2026.

• Competitive but not leading overall: Moonshot acknowledges that K3 still trails top proprietary systems like Claude Fable 5 and GPT-5.6 Sol in overall experience and some evaluations.

Why It Matters:
Kimi K3 narrows the gap between open and proprietary frontier models. Its strong coding, multimodal, and autonomous research capabilities give developers greater control over advanced AI, though running it still requires significant computing infrastructure.

Stock Updates

AI Brain Implant Helps Paralysed Man Regain Movement and Touch

“Double neural bypass” combines AI, brain electrodes, and electrical stimulation to reconnect movement and sensation

Researchers at the Feinstein Institutes report that an AI-powered brain implant helped Keith Thomas, who has complete tetraplegia, regain hand movement and touch. Some improvements lasted more than two years after stimulation stopped, suggesting lasting neuroplasticity.

Key Points:

• Movement restored: After training, Thomas could feed himself, drink from a cup, scratch his nose, and wipe his mouth using his own hand.

• Major strength gains: Over 35 weeks, his right arm became 86% stronger, while his left arm improved by 62%.

• Touch sensation returned: A technique called cortical mirroring helped restore feeling in a wrist that had been numb since his 2020 spinal cord injury.

• How it works: Five microelectrode arrays implanted in his brain detect movement intentions. AI translates those signals and triggers electrical stimulation of his forearm muscles.

• Artificial sensory feedback: Sensors in a 3D-printed brace stimulate the sensory cortex, allowing the participant to experience touch while handling objects.

• Long-lasting improvements: Many gains remained more than two years after stimulation ended, suggesting the nervous system was partly rewired rather than temporarily bypassed.

• Next steps: Researchers plan larger clinical trials and are exploring whether the system could also help people recovering from conditions such as stroke.

Why It Matters:
Restoring hand function is the top priority for many people with tetraplegia. While this result comes from a single participant, it suggests AI-powered brain-computer interfaces could help rebuild lasting nervous system connections, not just temporarily control muscles.

Take the TruAge™ Test and Get Your Personalized Blueprint for Health

What if a single test could unlock everything you need to know about your health? With TruDiagnostic, it can. From identifying your biological age to tracking key markers like inflammation, vitamins, and metabolic health, their reports are fully customized to you.

You’ll receive actionable steps based on your results, giving you the clarity and confidence to make impactful changes. One finger prick of blood, one life-changing report. Smarter With AI readers can save 20% today with code NEWSLETTERS20

AWS Billing Bug Shows Customers Owe Billions in False Charges

Amazon says inaccurate estimates do not reflect actual usage as engineers work on a fix

Amazon is fixing an AWS billing portal bug that displayed enormous estimated charges for services customers never used. Some accounts showed bills ranging from millions of dollars to nearly $2.5 billion, but Amazon says the figures are inaccurate and will not become real charges.

Key Points:

• False Billing Estimates: Some AWS customers saw estimated monthly bills worth millions or billions of dollars despite no corresponding cloud usage.

• Billing System Bug: Amazon linked the issue to a recent change in its billing computation subsystem.

• Rollback Failed: An initial attempt to reverse the change did not fix the problem, forcing AWS engineers to continue investigating.

• Customers Not Actually Charged: Amazon said the estimates do not reflect real usage or legitimate charges, meaning affected customers should not owe the displayed amounts.

• Account Impact Unclear: Amazon has not said whether any customer accounts were suspended or restricted because of the faulty billing alerts.

Why It Matters:
AWS billing data affects budgets, automated spending controls, and access to critical cloud systems. Even when the charges are not real, errors of this scale can cause panic, trigger internal alerts, and potentially disrupt businesses that rely on accurate cloud cost monitoring.

AI Highlights of the Week

• Chinese AI Just Broke the Pricing Curve

  Chamath Palihapitiya said cheaper AI models from Meta, Google, xAI, and China are closing the gap with premium models from OpenAI and Anthropic.

  He argued that for most business use cases, lower-cost models may now be good enough, putting pressure on expensive AI providers.

• Google Reportedly Delays Gemini 3.5 Pro

  Google has reportedly delayed Gemini 3.5 Pro after the model missed internal goals, especially in coding performance.

  The delay adds pressure as rivals like OpenAI and Anthropic continue pushing stronger AI models.

• Anthropic Launches Claude for Teachers

  Anthropic introduced Claude for Teachers, giving verified U.S. K-12 educators free access to premium Claude features.

  It includes teaching skills, standards-aligned curriculum support, and privacy protections for student data.

• Thinking Machines Releases Inkling Open-Weights Model

  Thinking Machines released Inkling, an open-weight multimodal AI model with 975B total parameters and 41B active.

  The model supports 1M-token context and is built for custom fine-tuning through the company’s Tinker platform.

Too Important to Miss

Canva Unveils AI Code Generator

NotebookLM is now Gemini Notebook

OpenAI's first gadget is a $230 keyboard

Netflix used AI in nearly 300 titles this year

Last Week’s Poll Result

• What does GPT-5.6 signal about OpenAI’s direction?

  AI is moving beyond chat → 41.18%

  Agents are becoming mainstream → 35.29%

  Work automation is accelerating → 23.53%

• Does Muse Spark 1.1 make Meta a serious AI competitor again?

  Maybe, but it needs real-world proof → 14.29%

  No, OpenAI and Anthropic still lead → 85.71%

• Can a cheaper model beat more expensive frontier models in developer workflows?

  Yes, efficiency wins → 46.15%

  Maybe, for most everyday tasks → 38.46%

  No, top capability matters most → 15.38%

Feedback

Know someone who may be interested?