Updated daily at 09:00 Beijing time
2026-10-16 · r1
Cutoff 2026-10-16T08:30:00+08:00
NVIDIA puts new open models into developers' hands, Google makes source links more visible, and Anthropic embeds external evaluators. Today: where models run, where information comes from, and who checks it.
Beyond capability, deployment conditions, traceable sources and external review shape whether AI products can deliver reliably.
Export PDF2026-10-15
NVIDIA released the 120B-parameter Super VL and Bolt 74B-A9B 24GB for local agents. Both are available now; the smaller Bolt E4B 4GB is scheduled for October 23.
FieldToFit: This advances the earlier launch preview. The 24GB target is useful for deployment planning; speed still depends on hardware, quantization and workload.
2026-10-15
AI Search will show inline links higher in answers more often and test a dedicated area for official destinations. Google also says its paid News AI licensing pilot now includes over 400 publications worldwide.
FieldToFit: Source visibility is becoming part of product competition. Link placement and licensing matter; publisher traffic gains still need evidence.
2026-10-15
Users can save items from multiple merchants across Search and Gemini, receive price-drop and restock alerts, and check out with Google Pay or on merchant sites. Merchants remain the sellers and provide confirmations and support.
FieldToFit: The update improves continuity across shopping sessions. It remains a shopper-controlled tool, not blanket authority for an agent to place orders.
2026-10-15
The stable release adds a test-failure explorer with search and analysis requests for up to 20 selected failures. Background Shell runs can outlive a turn and produce artifacts. The release notes list no known breaking changes.
FieldToFit: The focus shifts toward tracking engineering work. Existing failing tests offer a practical way to check search, recovery and artifact handoff.
2026-10-16
The stable release adds Flux3-image support, ImageSDRToHDR and ReferenceMultiLatent nodes, plus updated workflow templates. GitHub records publication at 06:57 Beijing time on October 16, before this edition's cutoff.
FieldToFit: This is a workflow compatibility update. Model capabilities, weight access and usage licences remain separate questions.
2026-10-15
Intel reports running the 816B-parameter model on four GPU D-series cards with 160GB each, alongside an INT4 weight-only path for Xeon. The article still describes GPU D-series as an upcoming platform.
FieldToFit: Memory capacity and low-bit software jointly shape deployability. This demonstration does not establish comparable end-to-end throughput.
2026-10-15
The European Commission lists the first AI in Science Enablers, based on voluntary in-kind pledges. NVIDIA announced open courseware and compute-access support; Google pledged $3 million in cloud credits or tokens, and Mistral pledged model and tooling resources.
FieldToFit: Research teams also need access routes, allocations and support. These pledges do not mean every lab has already received compute.
NVIDIA: European research compute and open courseware pledge
2026-10-15
Anthropic says six METR evaluators have access roughly comparable to its internal risk teams. METR can publish assessments without Anthropic veto, subject to redactions in predefined categories. Other evaluator partnerships retain more limited access.
FieldToFit: This is a voluntary governance arrangement. Watch the published findings, redaction transparency and follow-up actions; it is not a government regulatory conclusion.
Anthropic: Independent evaluator access and publication rights
3.5, 120B · 2026-10-15
Multimodal agent model; available now.
3.5, 74B-A9B 24GB · 2026-10-15
Local-agent model with 2-bit quantization; available now.
0.25.2 stable · 2026-10-15
Test-failure explorer and background Shell artifacts.
0.41.0 stable · 2026-10-16
Flux3-image support and new image nodes.
1.0.99-1 prerelease · 2026-10-16
One-off prompt-session folder trust and false context-exhaustion fixes.
US rollout · 2026-10-15
Cross-merchant saved items, alerts and user-directed checkout.
Regular US Eastern close. Close and daily change excerpted from Stock Analysis; excludes after-hours trading.
| Company | Close | Change |
|---|---|---|
| NVIDIA / NVDA | 217.23 | -1.84 |
| AMD / AMD | 562.23 | -1.5 |
| Intel / INTC | 95.37 | -2.8 |
| Microsoft / MSFT | 539.44 | 1.08 |
| Alphabet / GOOGL | 349.22 | 0.45 |
| Meta / META | 736.72 | -0.11 |
| Amazon / AMZN | 254.76 | 0.27 |
15:00 Beijing close. Close and daily change excerpted from Stock Analysis.
| Company | Close | Change |
|---|---|---|
| Cambricon / 688256 | 897.17 | -3.43 |
| iFlytek / 002230 | 36.3 | -0.03 |
| Hygon / 688041 | 193.82 | -3.79 |
Includes the closing auction; 16:10 is its market-wide end boundary. Figures are the unadjusted close and daily percentage change reported by Stock Analysis. The source does not specify a delay for these observations. These are dated closing values, not live quotes.
| Company | Close | Change |
|---|---|---|
| Baidu / 09888 | 78.35 | -4.51 |
| Alibaba / 09988 | 106.4 | -3.71 |
| Tencent / 00700 | 417.8 | -1.32 |
| SenseTime / 00020 | 1.14 | -4.2 |
| SMIC / 00981 | 52.15 | -3.96 |
Watch local-model performance on real tasks, traffic flowing from AI Search to original sites, and METR's published assessments. Released products, ongoing tests and resource pledges remain distinct stages.