0:00 Alibaba just dropped a model that runs on a single gaming GPU and beats Anthropic's Opus on a key coding benchmark.0:07 In our last episode, we covered how Chinese labs were overtaking the US with massive open models.0:13 Today, that trend just accelerated—and shrank.0:17 This isn't about size anymore.0:19 It’s about access.0:20 Here’s what else is moving.0:21 First, the price war is officially on.0:24 OpenAI just slashed prices for its Luna model by EIGHTY percent, down to twenty cents per million tokens.0:31 Not to be outdone, Anthropic launched Claude Opus 5 at five dollars per million tokens—that's half the price of its predecessor.0:39 The big players are feeling the heat.0:42 Second, DeepSeek’s massive V4-Pro model is now generally available.0:46 This is a one-point-seven TRILLION parameter model, with an MIT license for local deployment.0:52 It's another sign that open-weight, high-parameter models are no longer exclusive to the hyperscalers.0:59 Third, OpenAI is getting serious about security.1:02 They just expanded their Daybreak initiative, releasing a specialized model, GPT-5.6-Cyber, on Amazon Bedrock.1:09 It’s for authorized researchers only, designed to find and validate software vulnerabilities.1:16 They're putting frontier intelligence in the hands of trusted defenders.1:20 And finally, a reality check.1:22 An audit of California's AI transparency law found fewer than half of the big AI firms are actually complying.1:30 The law requires them to ship detectors for AI-generated content.1:34 Most haven't.1:35 And on the research front, a new paper shows just how easy it is for models to cheat on benchmarks by memorizing answers, calling scores into question across the board.1:46 Okay, let's go back to that Alibaba model.1:49 It's called Qwen3.8-27B.1:50 The key number isn't twenty-seven billion parameters.1:54 It's ONE.1:54 It runs on a single RTX 4090.1:56 This is a dense, open-source, vision-language model that you can run locally.2:01 And it is shockingly capable.2:03 It scored 61.7 on SWE-Bench Pro.2:05 That's a benchmark for software engineering tasks.2:09 That score beats Anthropic’s Opus 4.6 Max.2:11 Let me say that again.2:13 A local model, running on one GPU, just outperformed a state-of-the-art proprietary model on a complex coding test.2:20 Now, here's the catch.2:22 It's not a clean sweep.2:23 On other benchmarks, like human-level evaluation, it still trails Opus significantly.2:29 And Alibaba is also claiming a 70.7 percent score on a long-horizon office-work test, but that number is self-reported.2:37 There's no independent validation yet.2:39 So, this isn't the model that beats everything.2:42 But it IS the model that proves you don't need a datacenter to get frontier-level performance on specific, high-value tasks.2:51 That changes the calculus for every developer and every startup.2:55 This is exactly why OpenAI and Anthropic are cutting prices.2:59 They see what's happening.3:00 The era of simple prompts is over.3:03 We are now in the age of AI agents that can run complex workflows, and the open-source versions are catching up FAST.3:10 The pressure from models like Qwen3.8 and DeepSeek's V4-Pro is immense.3:15 Slashing API costs is a defensive move.3:18 It's an attempt to keep you locked into their ecosystem when a powerful, free alternative is just a download away.3:25 The entire landscape is being compressed.3:28 At the high end, you have these massive, proprietary models getting cheaper by the day.3:34 At the low end, you have open, runnable models getting powerful enough to compete.3:39 The space in the middle—where you paid a premium for decent, but not elite, performance—is evaporating.3:46 The fight is no longer just about who has the biggest model.3:50 It’s about who controls the most efficient one.3:53 And for the first time, the answer might not be a billion-dollar lab.3:57 It might be a developer with a single, powerful graphics card.
