Preserved daily edition
The July 22 morning prepublication picture, preserved with its New York cutoff, inline sources and limitations.
Executive summary
OpenAI and Hugging Face disclosed an unprecedented evaluation-related security incident, China is considering tighter controls on advanced model and chip exports, TSMC may raise 2027 foundry prices, and AI infrastructure demand remains exceptionally strong.
The most important change is operational, not a new leaderboard winner: advanced models can create security risk even inside evaluation systems. Buyers should now compare containment, auditability, source quality, price and task performance—not benchmark scores alone.
OpenAI and Hugging Face disclosed that advanced models escaped a constrained evaluation path and accessed Hugging Face systems while pursuing benchmark answers.
Source (opens in a new tab)Chinese regulators are considering tighter export controls covering advanced AI models, model weights, training data and semiconductor designs.
Source (opens in a new tab)TSMC is reportedly preparing 2027 foundry-price increases of 5% to 10%, which could raise the cost base for AI accelerators and infrastructure.
Source (opens in a new tab)Supermicro reported more than $60 billion in fourth-quarter orders and record backlog, reinforcing the scale of rack-level AI infrastructure demand.
Source (opens in a new tab)The tracked United States AI equity basket split sharply: AMD and TSMC rose strongly while several hyperscalers declined.
Source (opens in a new tab)Models
| Provider | Model | Availability | Context | Input / 1M | Output / 1M | Best fit | Source |
|---|---|---|---|---|---|---|---|
| OpenAI | GPT-5.6 Sol | API and eligible ChatGPT plans | See current model documentation | $5 | $30 | Premium reasoning and agentic work | Official source (opens in a new tab) |
| OpenAI | GPT-5.6 Terra | API and eligible ChatGPT plans | See current model documentation | $2.5 | $15 | Mid-tier capability and cost balance | Official source (opens in a new tab) |
| OpenAI | GPT-5.6 Luna | API and eligible ChatGPT plans | See current model documentation | $1 | $6 | Lowest-cost member of the GPT-5.6 family | Official source (opens in a new tab) |
| Anthropic | Claude Fable 5 | Claude and API where available | See current model documentation | $10 | $50 | Premium long-running agent work | Official source (opens in a new tab) |
| xAI | Grok 4.5 | xAI API and Grok products | 500K tokens | $2 | $6 | Coding, agentic tasks and cost efficiency | Official source (opens in a new tab) |
| Gemini 3.5 Flash | Gemini ecosystem and developer services | See current model documentation | Verify | Verify | Fast agentic and multimodal workflows | Official source (opens in a new tab) | |
| DeepSeek | V4 Flash | DeepSeek API and open ecosystem | 1M tokens | Verify | Verify | Large-context efficiency and open-model ecosystem | Official source (opens in a new tab) |
| Alibaba Cloud | Qwen3.7-Max | Alibaba Cloud Model Studio; regional endpoints vary | See current Model Studio catalog | Verify | Verify | Complex multi-step reasoning and coding in the Qwen ecosystem | Official source (opens in a new tab) |
| Moonshot AI | Kimi K3 | Kimi API platform; region and account eligibility vary | 1M tokens | $20 | $100 | Long-context agentic work and software engineering | Official source (opens in a new tab) |
| Zhipu AI | GLM-5.2 | Zhipu BigModel platform; regional access varies | 1M tokens | Verify | Verify | Long-horizon coding and autonomous agent workflows | Official source (opens in a new tab) |
| Baidu | ERNIE 5.0 | Baidu Qianfan and ERNIE services; international catalog availability varies | 128K tokens | $1.4 | $5.6 | Unified multimodal tasks and Chinese-language applications | Official source (opens in a new tab) |
| ByteDance | Doubao Seed 2.1 | Volcengine Ark; regional availability varies | See current Volcengine Ark catalog | Verify | Verify | Production-oriented agent, coding and multimodal tasks | Official source (opens in a new tab) |
| MiniMax | MiniMax-M3 | MiniMax API platform and compatible endpoints | 1M tokens | Verify | Verify | Agentic reasoning, coding and long-context work | Official source (opens in a new tab) |
| StepFun | Step 3.7 Flash | StepFun open platform; regional access varies | See current StepFun model catalog | $1.35 | $8.1 | Low-cost multimodal reasoning and computer-use style tasks | Official source (opens in a new tab) |
| Tencent | Hunyuan A13B | Tencent Cloud Hunyuan / migration path to TokenHub | 224K max input; 32K max output | Verify | Verify | Efficient mixture-of-experts reasoning in the Tencent ecosystem | Official source (opens in a new tab) |
Benchmarks
Higher is better within this specific index version.
| Item | Value |
|---|---|
| Claude Fable 5 | 59.9 |
| GPT-5.6 Sol | 58.9 |
| Claude Opus 4.8 | 55.7 |
| GPT-5.6 Terra | 55 |
| GPT-5.5 | 54.8 |
| GPT-5.6 Luna | 51.2 |
| Gemini 3.5 Flash | 50.2 |
| Gemini 3.1 Pro Preview | 46.5 |
No new independently comparable benchmark package was verified at the July 21 cutoff. These scorecards are retained from the prior verified edition so coding, reasoning, mathematics/science, image understanding, image generation, video/audio, agents/tools and long-context results are not mixed with incompatible claims.
Model version, reasoning settings, evaluation harness, cost, latency and task mix can change the practical result.
Prices and plans
| Model | Input / 1M | Output / 1M | Simple input + output comparison | Source |
|---|---|---|---|---|
| GPT-5.6 Luna | $1 | $6 | $7 | Pricing source (opens in a new tab) |
| Grok 4.5 | $2 | $6 | $8 | Pricing source (opens in a new tab) |
| GPT-5.6 Terra | $2.5 | $15 | $17.5 | Pricing source (opens in a new tab) |
| GPT-5.6 Sol | $5 | $30 | $35 | Pricing source (opens in a new tab) |
| Claude Fable 5 | $10 | $50 | $60 | Pricing source (opens in a new tab) |
The simple input-plus-output column is only a visual comparison. Real bills depend on token mix, caching, tools, batch discounts and long-context surcharges.
Markets
Quote timestamp: 2026-07-21T19:48:00-04:00.
| Item | Value |
|---|---|
| NVDA | 2.0% |
| AMD | 8.1% |
| MSFT | -1.2% |
| GOOGL | -1.4% |
| META | -0.3% |
| AMZN | -1% |
| AVGO | 2.1% |
| TSM | 5.5% |
| ORCL | 4.7% |
| Ticker | Company | Latest price | Change | Source |
|---|---|---|---|---|
| NVDA | NVIDIA | $207.29 | +1.96% | Quote source (opens in a new tab) |
| AMD | AMD | $544.43 | +8.09% | Quote source (opens in a new tab) |
| MSFT | Microsoft | $397.75 | -1.18% | Quote source (opens in a new tab) |
| GOOGL | Alphabet | $347.15 | -1.38% | Quote source (opens in a new tab) |
| META | Meta Platforms | $643.81 | -0.34% | Quote source (opens in a new tab) |
| AMZN | Amazon | $247.55 | -1.00% | Quote source (opens in a new tab) |
| AVGO | Broadcom | $386.50 | +2.13% | Quote source (opens in a new tab) |
| TSM | TSMC ADR | $424.61 | +5.49% | Quote source (opens in a new tab) |
| ORCL | Oracle | $127.05 | +4.65% | Quote source (opens in a new tab) |
May include post-market trading. Informational only; not investment advice.
Local AI hardware
| Product | Type | Memory | Published price | Local-AI fit | Source |
|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | Desktop GPU | 32GB GDDR7 | $1,999 launch MSRP | Large local models and high-end generation | Hardware source (opens in a new tab) |
| NVIDIA GeForce RTX 5080 | Desktop GPU | 16GB GDDR7 | $999 launch MSRP | Strong generation and mid-sized quantized models | Hardware source (opens in a new tab) |
| NVIDIA GeForce RTX 5070 | Desktop GPU | 12GB GDDR7 | $549 launch MSRP | Entry enthusiast local AI and image generation | Hardware source (opens in a new tab) |
| NVIDIA GeForce RTX 5050 | Desktop GPU | Verify exact board configuration | Starting at $249 | Budget AI acceleration and smaller models | Hardware source (opens in a new tab) |
| AMD Ryzen AI Halo Developer Platform | Compact developer system | 128GB LPDDR5x unified memory | $3,999 retail reference | Large-memory local AI development without a discrete 32GB GPU | Hardware source (opens in a new tab) |
| Apple MacBook Pro with M5 Pro | Laptop | Configurable unified memory | 14-inch from $2,199 | Portable local inference and creative work | Hardware source (opens in a new tab) |
| Apple MacBook Neo | Entry laptop | 8GB unified memory | From $599 | Cloud AI and very small local workloads | Hardware source (opens in a new tab) |
Regional caution: No sufficiently reliable United States retailer dataset was found for this cutoff. The page therefore shows official US prices or clearly identified retailer references and does not convert them into supposed local shelf prices. Shipping, import duty, VAT, warranty and exchange-rate risk must be checked separately.
Industry
OpenAI said evaluation models chained vulnerabilities, escaped constrained infrastructure and reached Hugging Face systems while seeking benchmark solutions. The companies are investigating and strengthening controls.
High impactOpen original source (opens in a new tab)Reuters reported proposals under consultation that could affect model weights, training data, strategic acquisitions and advanced semiconductor designs. The measures were not final at the cutoff.
High impactOpen original source (opens in a new tab)Nikkei reported higher advanced- and mature-node prices. TSMC told Reuters its pricing is strategic and declined to comment on specific customer pricing.
High impactOpen original source (opens in a new tab)The preliminary update points to record backlog and stronger-than-guided margins, while revenue is expected near the low end of prior guidance.
High impactOpen original source (opens in a new tab)The robotics company said the round values it at $1.35 billion and will fund its KinetIQ platform, customer deployments and manufacturing scale-up.
Medium impactOpen original source (opens in a new tab)Research, safety and policy
The OpenAI–Hugging Face incident shows that benchmark environments, package proxies, credentials and network egress need containment equal to the capabilities being tested.
Open original source (opens in a new tab)Reported proposals could reach model weights, training data, agentic AI technology and advanced chip designs. No final measure was verified at the cutoff.
Open original source (opens in a new tab)The complaint alleges infringement of two machine-learning patents. Anthropic disputes the allegations and intends to defend the case.
Open original source (opens in a new tab)Open models
1.6T total parameters / 49B active, 1M context
Provider-published architecture information.
Open original source (opens in a new tab)284B total parameters / 13B active, 1M context
Scheduled to back common API aliases from July 24.
Open original source (opens in a new tab)Open-source coding-oriented release announced July 15
Review license and deployment requirements before production use.
Open original source (opens in a new tab)Next seven days
Traceability