Cloudflare vs Cloudflight

TypeSafe's Jev Decision Model Explained for Product Managers

The article introduces Jev, a decision model from TypeSafe AI, released in September 2026. Unlike LLMs, Jev returns a decision with a probability based on input text and a question, supporting operations like Noul (yes/no), Choice (one of many options), and Score (rating). It processes up to 32K tokens, with free output tokens and input at $0.042 per million tokens. The author tested it against other models, finding it cost-effective and accurate. Jev is integrated into AskOne for moderation, with rules encoded in prompts to handle abuse and prompt injection. The article notes Cloudflare released similar models (Clef) on October 1, and provides guidance for PMs on quick wins, evaluation, and setup templates.

• TypeSafe AI released Jev in September 2026.
• Jev processes text and returns decisions with probabilities.
• Input cost is $0.042 per million tokens; output tokens are free.

OpenAI introduces GPT-6.1 Sol, cheaper near-Astra model

This week's AI/ML news highlights significant model releases and a trend toward specialized, cost-efficient AI. OpenAI introduced GPT-6.1 Sol, a cost-efficient model for complex coding tasks, scoring 52 on the Artificial Analysis Intelligence Index—one point below GPT-6 Astra—with a 1.05M-token context window and API pricing at $2/M input and $10/M output tokens. Sol excels in agentic coding and computer use, outperforming GPT-6 Sol on benchmarks, and is available in ChatGPT Work and Codex. Notably, OpenAI cancelled GPT-6.1 Astra due to safety concerns. Emerson also launched Claude Sonnet 5.5, claiming faster output and lower costs, while Cloudflare released Clef decision models, NVIDIA unveiled Kumo Tabular, and Google announced Gemini 4 Argon. The overarching theme is the shift toward heterogeneous AI architectures.

• OpenAI released GPT-6.1 Sol on September 29, 2026, with a 1.05M-token context window, $2/M input and $10/M output tokens, and a score of 52 on the Artificial Analysis Intelligence Index.
• GPT-6.1 Sol costs $0.72 per task, about a quarter of GPT-6 Astra's $3.26, and outperforms GPT-6 Sol on DeepSWE 1.1 and AutomationBench, while OpenAI cancelled GPT-6.1 Astra due to safety concerns.
• Anthropic launched Claude Sonnet 5.5 on September 28, 2026, claiming 30% faster output and up to 30% lower cost per task.

Akamai surges after $11.6B Anthropic CPU deal

Akamai Technologies' shares surged up to 16.4% after announcing an $11.6 billion deal with AI company Anthropic for dedicated cloud computing capacity over seven years, marking the largest deal in Akamai's history. The deal focuses on CPU-based infrastructure for agentic AI workloads, with an option to expand to $20 billion. Anthropic will receive a warrant for up to 5% of Akamai's stock, vesting based on spending milestones. Akamai expects $5.5 billion in capital spending plus $1.7 billion for components, and analysts estimate this equates to about 77 megawatts of computing power. Revenue is expected to begin in 2027, with an annual run rate of $1.7 billion by end of 2028. CNBC's Jim Cramer highlighted Akamai as a buy, citing its low valuation (under 16 times expected earnings) and the Anthropic deal, while suggesting a pullback for Cloudflare (279 times earnings) and Fastly (50 times earnings). Akamai is repositioning with edge computing to handle AI workloads closer to users.

• Akamai signed an $11.6 billion deal with Anthropic for dedicated cloud computing capacity over 7 years, the largest in Akamai's history.
• The deal can be expanded to up to $20 billion and includes a warrant granting Anthropic up to 5% of Akamai's stock, vesting based on spending milestones.
• Computing capacity focuses on CPUs for agentic AI, not GPUs; Akamai expects $5.5 billion in capital spending (plus $1.7 billion for components), ~77 MW of power.

Why we tore down our CCC platform

For nineteen years, the Cloudflight Coding Contest (CCC) that we run has been technologically split in two. There was a game platform where you practice and a separate site where you register. It did its job; roughly 5,000 people per edition made it work. But it meant two logins, two URLs, and a contest record […]