BTA Labs

@bridgemindai My Codex limit in plus plan was gone after ONE single day with ONE agent at high effort.

GPT 5.6 usage is feels unlimited in Codex right now. I ran 30 agents at once. It used 2% of my weekly limit. I have 5 usage resets sitting in the bank on top of that. I don't think I could hit my limit if it was my job to.

Kimi K3 lanching tomorrow! Will it beat GLM-5.2?

@bijanbowen Thanks for the nice video! I honestyly prefer your videos over benchmarks

@ApplyWiseAi Imagine having the full power of Mythos 5 without any restrictions..🔥

@BTA_labs benchmarks are unreliable now but fable 5 was different. that one felt real.

On June 9, Anthropic dropped Fable 5. And for a moment, the entire AI race felt decided. Fable 5 wasn't just another model. It was the first Mythos class engine released to the public, and it obliterated nearly every benchmark in sight. Long context reasoning, autonomous coding, multi step research, Fable 5 handled things no other model could touch. The second best model? Also Anthropic's. Opus 4.8. They weren't just winning. They were lapping the field. Then, three days later, it vanished. Gone. Globally. No warning, no graceful sunset, no replacement model waiting in the wings. Here's why: researchers figured out how to prompt Fable into detecting software vulnerabilities at a level that set off national security alarm bells. The US government hit it with export controls, and Anthropic couldn't verify every user's nationality in real time, so they had to pull the plug worldwide. It came back July 1, but different. Heavier cybersecurity classifiers, government pre release testing, data retention on Mythos class traffic. And here's the part barely anyone is talking about: Anthropic has been quietly moving the goalposts on availability ever since. Fable 5 was supposed to leave subscription plans and shift to paid credits starting July 7. Anthropic extended it to July 12. Then, two days before that deadline, they extended it again to July 19. Same weekly usage cap, same "we're working on restoring it permanently" language, three times in a row. Anthropic says the extensions are because of compute shortages, not a policy reversal. But notice the timing. Each extension has landed right as a competitor dropped something big. GPT 5.6 launched July 9. Grok 4.5 launched July 8. The same week both hit, Anthropic extended free Fable access again. Some people online are calling it exactly what it looks like: a retention play disguised as generosity. So now the pattern looks like this. A government can shut your model off entirely with no warning. And even after it's restored, the company can't tell you for more than a week at a time whether you'll still have free access to it next Tuesday. That's not a stable foundation for anything you're building a business on. While all this was happening, the rest of the industry didn't wait around. OpenAI shipped GPT 5.6 in three tiers, Sol, Terra, and Luna, and just days later removed the 5 hour usage cap entirely for Plus, Pro, and Business, resetting quotas because demand doubled. SpaceXAI shipped Grok 4.5, trained with Cursor, now ranking fourth on the intelligence index. Meta pushed Muse Spark 1.1 into a real paid API. On the open side, GLM 5.2 beat many good models on SWE Bench Pro, and Qwen 3.6 jumped two generations, runnable on consumer hardware. Every one of these companies is playing a different game with different incentives, and none of those incentives are about your stability. OpenAI loosens limits to win usage share. Anthropic extends access week by week to avoid losing users to OpenAI. A government can pull a model with zero notice if it decides the capability is a national security risk. None of that is under your control, and none of it will ever be predictable. So yes, local models aren't there yet. Qwen 3.6, Gemma 4, GLM 5.2, none of them fully replace Fable 5 today. But look at the actual trajectory instead of the current gap. A year ago, the idea of running anything close to frontier reasoning on your own hardware sounded like a joke. Local models were toys, good for basic autocomplete and simple chat, nothing more. Today, a 27B dense model runs comfortably on a single high end laptop GPU and beats models three times its size from just two generations back on real coding benchmarks. Quantization got better. Distillation got better. The tooling around running these models locally, llama.cpp, vLLM, LM Studio, went from hobbyist scripts to production grade infrastructure almost overnight. The distance between "open weight toy" and "usable daily driver" collapsed in about eighteen months. Now project that same curve forward. If open weight labs keep closing the gap at anything close to the rate they closed it over the last year, a model with genuine Mythos level reasoning, not just decent chat, but real long context planning and agentic coding, running fully offline on a normal laptop, isn't a moonshot. It's the next predictable step. The compute cost to train frontier capability keeps dropping. The techniques to compress that capability into something that fits on consumer silicon keep improving. Nothing about the last eighteen months suggests that curve is slowing down, and every release cycle from Qwen, DeepSeek, GLM, and Gemma has been compressing the timeline further, not stretching it out. That's not wishful thinking. That's just where the math is pointing. Use the frontier models hard right now. They're genuinely incredible for the hardest reasoning and research problems, and pretending otherwise would be dishonest. But if your business depends on a subscription that gets extended one week at a time by a company managing its own compute shortage, you don't have a tech stack. You have a lease that renews at the landlord's mercy, and the landlord is currently improvising. Keep your core execution local. Keep fallbacks running. Build so that no government, no company, no rolling deadline can quietly decide how you work. Because within a year, the tradeoff between "use the frontier API" and "run it yourself" is going to look completely different than it does today, and the people who already built the muscle to run local will be the only ones ready for it.

@BLCNYY @thsottiaux Dont have it either

Where is my reset? 🤔 @thsottiaux

@tOMPSON Glaubst du, dass FSD in Deutschland sofort aktiviert werden kann, wenn man von einem anderen Land kommt (kein FSD)?

Ich warte schon zu lange auf den Tag, an dem FSD bei uns genehmigt wird. Ich habe schon zu oft gehofft, dass es bald soweit ist. Trotzdem hoffe ich, dass da jetzt wirklich was dran ist und dass Österreich dann gleich mitzieht #tesla #fsd 🤞🏻

@MattGriscom Yes, I think their summer break starts in a few weeks, so something should happen within the next few days.

@BTA_labs That would be awesome!

FSD in Germany will be approved in the coming days!

@TSLA_inside_ Something has to happen before they go to vacation this month😁

The German Minister said that something will still happen this month. I hope that means this week. 🤞🇩🇪 Everyone in Germany wants FSD (Supervised). And let’s not forget: if Germany accepts Tesla FSD through the Dutch RDW approval under Article 39, it could trigger a chain reaction, with other countries following the same path. The momentum is building. 🚗⚡

@teslaeurope Italy, Germany, France, Spain soon?

Over 50 million kilometers driven on FSD Supervised by our customers in the Netherlands, Estonia, Belgium, Lithuania & Denmark 🤖 🇳🇱 🇪🇪 🇧🇪 🇱🇹 🇩🇰

@julian_raich Diesen Monat aber bestimmt

@BTA_labs Nein kommt nicht

@vibecademyai Grok is underrated

Fable 5 > GPT 5.6 > Opus 4.8 > GLM 5.2 > Grok 4.5

@bridgemindai Fable extended!

Today is the LAST DAY Fable 5 is on Claude subscriptions. So let me be direct with Anthropic: GPT 5.6 is way better than Opus 4.8. Without Fable 5, there is nothing on your subscription worth paying for. I have 4 Claude Max subscriptions. $800 a month. If Fable 5 is not brought back, I cancel all of them immediately. Your move.

Claude Fable 5 extended until Opus 5 comes out it seems...

@Bhavani_00007 Fable 5 only would also be ok...

Anthropic if compute is the real problem, don't pull Fable 5 from the subscription cut Sonnet 5 instead > nobody in my circle uses it > it eats tokens like crazy > and brings no real advantage over Opus in quality or cost just keep Opus and Fable 5 in the plan that is literally all we need 🙏

@kimmonismus Do you think Opus 5 will be as good as Fable 5, and that Anthropic will drop Fable 5.1 when they have enough capacity?

Today marks the end of Fable 5's subscription plan, presumably for an extended period. While Anthropic has made it clear they intend to keep Fable in the subscription plan in the future, they haven't specified a date. GPT-5.6 Sol is a good alternative for now, although there are certainly significant differences. But as I've mentioned several times, the rates with 5.6 are enormous, so its use remains somewhat limited at present. In any case, 5.6 was a major release that undoubtedly gave OpenAI a significant boost compared to Anthropic. Now the question is how Anthropic will handle this. My guess: They will release Opus 5 very soon as a cheaper alternative to Fable 5, hoping that this will appease the public. Sonnet 5 is hardly worth mentioning after its release, and I haven't used it myself. Therefore, I don't believe we'll see Fable 5 back in the plan anytime soon, but I'm happy to be proven wrong. (P.S.: @thsottiaux , please reset the rates 👉👈)

@bridgemindai @vibecademyai And Grok?

@vibecademyai I agree

Fable 5 > GPT 5.6 > Opus 4.8

@alex_avoigt I really hope an approval of Germany and Italy comes within the nex few weks

Good FSD news from 🇮🇹 Italy, they are im the process to review and hopefully decide without further delay.

@bridgebench But mich faster and more reliable

Grok 4.5 is on the same level as GLM 5.2.