cheaty

fable was very good at webslop can i have my shareholder value delivering model back please purely for shareholder value generation definitely not going to generate tons of traces to finetune qwen and gemma on for cybersec... wouldn't even think of such a thing

Fable being the enriched uranium of LLMs is a pretty good definition

@jxnlco please let me disable skills/only have skill names in context so i don't have to manually add a whole bunch of shit under disabled skills in the codex config.toml manually claude code does this well, you can toggle whether you want just the name or full description in context

codex users! two things i want feedback on: are plugins actually making codex better? which ones, and what still feels broken? how are you using codex as a team? drop examples and i’ll organize them for the team!

and so the underground token blackmarkets also grow exponentially in the singularity interesting

@emilstridell @scaling01 rip i thought m2.7 was 280b total

@cheatyyyy @scaling01 50% more params? it’s almost 100% more params brochacho haha

it's too small I smell benchmaxxing

@abdurrahmanregi @Zai_org opencode models --refresh

@Zai_org zai-coding-plan in opencode does not recognize glm-5.2

Intelligence should be open, accessible, and ready to build with, empowering every developer, everywhere. GLM-5.2 is now available to all GLM Coding Plan users, including Lite, Pro, Max, and Team plans. docs.z.ai/devpack/latest… As our new flagship model, GLM-5.2 delivers powerful coding capabilities, usable 1M-context support, and continued strengths in long-horizon tasks. API and Chatbot services will launch next week. The model will also be officially open-sourced next week under the MIT License. The future of AI is open, and it belongs to the people.

@_xjdr early impressions?

Lets see how this stacks up . high hopes

uh oh

@VictorTaelin this is like having my first born taken away from me

what did you do this time pliny

@thdxr hunter2

what's your password?

@synthwavedd @EpochAIResearch scaling is all you need

🚨 Claude Fable 5 scores 87.8% on FrontierMath Tier 4 v2, putting it in 1st place (ahead of DeepMind) and marking a 31.7 percentage point jump over Opus 4.8! On Tiers 1-3, it scored 87%, beating GPT-5.5 to also take 1st place and jumping 7 percentage points over Opus.

@TotallyNotParth @teortaxesTex you can start here because i have a bunch of hardcoded literal live examples of attack surfaces on real projects ive found as references in mine lmfao these 20 should be plenty to get started, deepseek v4 pro is happy to help skills.sh/yaklang/hack-s…

@cheatyyyy @teortaxesTex mind sharing said skills?

would be really strange of them to say this and then NOT begin faster iteration cycles. I still expect V4.1 in June

@RhysSullivan that's a minimum 9 dollar call, uncached

fable resuming the 900k token chat

@Sauers_ at 1m? i havent noticed anything at 400k+

Fable has end-of-context-window anxiety, like Claude Sonnet 4.5

@teortaxesTex i would love to read a paper on why claudespeak is easy for me to read but any sort slop/linkedin speak horribly itches my mind

But on the plus side, Fable has the best understanding yet of what's wrong with it. It is correct: what we hate is not "AI slop". It's very human slop indeed.

@kr0der 4x 5 hour limits are worth it

might be time to actually get the $200 claude plan to use more fable 💀