00:06GPT 5.6 just leaked and the receipts are everywhere. Multiple devs spotted a model
00:12called GPT 5.6 in OpenAI's codex backend logs. Internal codenames are floating around.
00:18Iris Alpha, Ember Alpha, and Beacon Alpha. That's three checkpoints being A-B tested
00:22at once. Here's the kicker. 1.5 million token context window. GPT 5.5 capped at 1.05 million.
00:29That's a 43% jump in one release. Polymarket has GPT 5.6 dropping before June 30th at over 85
00:36%
00:36probability. Insiders say it'll match Claude Mythos at 2 to 3 times cheaper. And about Mythos,
00:42Anthropix Secret Monster leads 17 of 18 benchmarks. 93.9% SWE Bench Verified. 97.6% on the USA
00:50Math
00:51Olympiad. Saturated SciBench at 100%. Mozilla turned Mythos loose on Firefox. It found 231
00:57zero-day vulnerabilities. 10 times more than the previous Claude model. Microsoft accidentally
01:02leaked the training compute on a slide. 6.1 times 10 to the 27 flops. That's trillions of parameters.
01:09Subscribe to GPT AI Clips. The free guide drops the full benchmark table and release dates in the
01:14description.
Comments