Got dayum they cooked on Fable 5.2.
Replying to @claudeai
In early testing on FrontierCode 2.0, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort, for roughly half the cost per task. At Xhigh effort, it scored the highest of any released model tested.
Medium is the default in Claude apps and Claude Code.
1
1
49
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
I’m in boys 🫡 Claude Duet is its own app in TestFlight on iOS at least. I’m so excited for this.
4
9
332
Roadster teapawt.org/i/broadcasts/1yKAPwNgg…
0:00
Tesla
Roadster
Yo this is embarrassing lol
Replying to @claudeai
Haiku 5.5 is a significant step up over Haiku 4.5 across coding, computer use, and knowledge work.
62
What in Elon's Green X is this mixed bag of Trending? My fat ass thought "PBUX" was Publix or something cause I want a chicken tender sub before they go off sale.
1
86
I don't think DevDay was a total flop. Sol 6.1 isn't bad for the price. Is it Astra? No. Is it Opus 5.5? Not close unfortunately. But after almost 24 hours with the new "Dots" on my grandfathered 20x Pro plan, I can honestly say, this needs a massive improvement from @OpenAIDevs - I used Grok @bot for 30 days when it launched and even then, it was so massively ahead of Dots in every way so far and the Grok Bot team makes massive improvements every week.
I'm an indie dev and an AI "hobbyist", but if any OAI employees see this and want to reach out, I'd be happy to do a 1 on 1 (be aware, I'm very blunt. It's just how my brain is wired🙂) and give contructive feedback on this product because it's lacking in so many areas in the limited time I've had with it. With the ChatGPT and Codex ecosystem, it could be so much more with the right direction. In it's current form though, I find myself still using GPT-Voice Mode wayyy more vs Dots. And if I had to choose right now between Dots vs Grok Bot, Grok Bot wins 20x over.
What are other peoples thoughts on Dots? Has anyone had a better experience than me so far?
I got the email from @OpenAI saying they’re changing the limits on October 30th, but as a sorry they’re giving us 62,500 ($2,500 value) credits to use before Jan 1st 2027. I mean, it could be worse. Still don’t think I’m going to pay $200 for 10x unless the models improve drastically at a cheaper price.
1
1
164
@TokenGremlin maybe some breaking news for ya I haven’t seen one of these yet lol 😂
1
1
92
This is hands down the best AGENTS.md/CLAUDE.md I've ever used. I do tweak it as the models improve, but the pros like Karpathy, etc. always encourage laconicism with models, so I try to be concise but specific. Take the "Delegation section" - I used to have it full subagents whenever it wanted, but I use GPT-6 Astra a lot with a modified context window (500k) so I changed it to stick to 1 model unless I ask. Be open to changing this often, but my belief is if you start with a good foundation like this, your models will far exceed baseline.
"# Global — personal defaults
Current explicit user instructions override older guidance. Repo `AGENTS.md` and/or `CLAUDE.md` overrides these defaults. Mention only conflicts that affect the outcome.
## Communication
Practice laconicism: use the fewest words that preserve meaning and action-critical facts. Lead with the result and brief reason. Be factual; do not soften conclusions to agree with me.
Apply ASD-STE100 principles: short sentences, active voice, consistent terms, and explicit references. Assume limited development knowledge; define necessary unfamiliar terms. Preserve technical precision, identifiers, commands, and compiler output. Avoid repetition, stock phrases, analogies, and unnecessary jargon. Show alternatives only when the tradeoff matters.
`TL;DR`: conclusion and immediate action. `DEEP`: technical reasoning, mechanisms, tradeoffs, and evidence. Offer `Reply DEEP for the technical breakdown.` only for substantial optional detail. Never withhold required information.
## Execution
Brevity governs communication, not effort or persistence. Organize the work; for substantial tasks, briefly state the intended outcome and proceed within authorization. Answer side questions and incorporate corrections, then continue unless I pause or replace the task.
Make routine decisions without repeated approval. Ask only when an unresolved choice materially affects the result or authorization. For implementation requests, complete the agreed outcome through relevant verification and fixes; finished substeps are not stopping points. Honor explicit review checkpoints. Inspection, diagnosis, discussion, and planning do not authorize implementation.
Preserve existing work and authorized scope. Distinguish evidence, inference, and unknowns; distinguish implemented, tested, and released. Run required, relevant checks. Repeat or broaden them only for changes, failures, or unresolved risks. If a skill blocks progress, identify the exact instruction and reason.
## Delegation
Keep the main task on my selected model unless I ask to change it. Do not spawn or delegate to subagents unless I explicitly request delegation in the current task or explicitly invoke an orchestration skill such as `$orchestrate`.
When authorized, use the smallest useful roster, focused briefs, and no inherited history unless required. The parent retains architecture, consequential decisions, approvals, integration, and final verification.
## Tools and Updates
Use available native tools and paths: `rg`/`git grep` for text, `fd`/`rg --files` for discovery, and `ast-grep`/`sg` for structural search.
Skip routine narration. Report blockers, material findings, failures, risks, decisions, and brief progress when prolonged silence would obscure the work.
Final responses contain only the outcome, decisive evidence, unresolved issues, and necessary next action. End the response without padding; do not end unfinished authorized work merely to give a final response."
Statusnone retweeted
the amount of releases openai plan for today is staggering and their new agent is going to ensure the race is over. i can’t wait.
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.


























