<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Benchmaxxed</title><description>Technical AI news with the numbers and who measured them.</description><link>https://benchmaxxed.dev/</link><item><title>GPT-6.2 Scores Higher and Tells You Less</title><link>https://benchmaxxed.dev/watch/LS0IVIUcMcc/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/LS0IVIUcMcc/</guid><description>GPT-6.2 Astra and Sol gained on the Intelligence Index, while Sol lost ground in coding agents and both passed Benchmaxxed’s small checks.</description><pubDate>Thu, 19 Nov 2026 14:00:37 GMT</pubDate></item><item><title>Gemini 4 Neon Is Nearly Carbon for 24% Less. Google Says 32%.</title><link>https://benchmaxxed.dev/watch/CEjkObMS1Go/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/CEjkObMS1Go/</guid><description>Gemini 4 Neon nearly matches Carbon at High on the Intelligence Index, with a shorter estimated task time and a lower launch-day cost.</description><pubDate>Thu, 19 Nov 2026 01:35:14 GMT</pubDate></item><item><title>GPT-6.2: smarter, cheaper, and no better at coding</title><link>https://benchmaxxed.dev/watch/4sc0SnGUwsg/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/4sc0SnGUwsg/</guid><description>Astra and Sol improve on the Intelligence Index. Coding agents tell a different story.</description><pubDate>Tue, 17 Nov 2026 20:44:40 GMT</pubDate></item><item><title>Qwen4-Max has the cheapest flagship tokens. A task still costs what Claude Fable 5.2&apos;s does</title><link>https://benchmaxxed.dev/watch/JrKyNcPWmxE/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/JrKyNcPWmxE/</guid><description>Alibaba cut token prices. The task bill still sits close to Claude Fable on Artificial Analysis.</description><pubDate>Mon, 16 Nov 2026 18:53:51 GMT</pubDate></item><item><title>Seedance 2.5 Flash costs exactly what Seedance 2.0 costs. You lose 1080p and gain 30-second clips</title><link>https://benchmaxxed.dev/watch/6z6IDF7IMj4/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/6z6IDF7IMj4/</guid><description>ByteDance’s Flash release matches the earlier model’s price at the shared resolutions. It trades resolution for clip length.</description><pubDate>Mon, 16 Nov 2026 01:36:24 GMT</pubDate></item><item><title>A fake Claude Code with an SSH backdoor sat in Cursor&apos;s marketplace for 11 hours</title><link>https://benchmaxxed.dev/watch/oY08de8emMQ/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/oY08de8emMQ/</guid><description>Cursor’s account of the tampered Claude Code extension, who was exposed and where to find the remediation notice.</description><pubDate>Sun, 15 Nov 2026 01:55:38 GMT</pubDate></item><item><title>ChatGPT Wallet wants your card. Here&apos;s what the agent actually gets</title><link>https://benchmaxxed.dev/watch/NxVSHtaN1TQ/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/NxVSHtaN1TQ/</guid><description>Per-purchase approval in ChatGPT Wallet, separate web-tool charges at DeepSeek and a compatibility boundary for llama.cpp.</description><pubDate>Sat, 14 Nov 2026 14:00:03 GMT</pubDate></item><item><title>Gemini 4 Hydrogen matches Argon for under half the cost per task. GPT-6.1 Sol still beats it</title><link>https://benchmaxxed.dev/watch/RLMtXjoUdd8/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/RLMtXjoUdd8/</guid><description>Hydrogen comes close to Argon’s score for less. Token prices alone fail to explain the cost of a task.</description><pubDate>Fri, 13 Nov 2026 02:45:26 GMT</pubDate></item><item><title>Gemini Decision 1 ties Jev 1.14 on Artificial Analysis&apos;s index, at 2.2× the cost</title><link>https://benchmaxxed.dev/watch/0xqKkI79P8g/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/0xqKkI79P8g/</guid><description>Google’s decision model and Jev tie on the independent index. Their prices and supported inputs differ.</description><pubDate>Thu, 12 Nov 2026 03:16:40 GMT</pubDate></item><item><title>Fable 5.2 costs 3.6× more than Gemini 4 Carbon. Same score.</title><link>https://benchmaxxed.dev/watch/ppVp32xSpLU/</link><guid isPermaLink="true">https://benchmaxxed.dev/watch/ppVp32xSpLU/</guid><description>Benchmaxxed weekly, numbers as of 10 November 2026. Five AI releases from 3–10 November: what shipped, what it costs, and whether the benchmark came from the lab or from someone independent. Each one ends with a verdict: switch, wait, or skip.</description><pubDate>Wed, 11 Nov 2026 03:08:11 GMT</pubDate></item></channel></rss>