Sabitlenmiş Tweet
Developed Life 🗿💹🧲
4.7K posts
codex tip: 🌕Luna has its BEST mode hidden by default. Go to Settings -> Configuration -> turn on Max.
Congratulations, you just unlocked the world's best model today
English
@avrldotdev If you use a personal subscription (not enterprise) you get far more usage when compared to the same cursor equivalent subscription.
In addition to certain models being better in their native harness.
English
now that GPT 5.6 Luna is WAY cheaper, are you guys running Luna?
would 5.6 Luna xhigh be comparable in quality over GPT 5.6 Sol med in exchange for faster speeds?
or is Luna still subagent-only territory?
English
Inkling Small from @thinkymachines on ARC-AGI (Verified):
- ARC-AGI-2: 40.1%, $0.23/task
- ARC-AGI-1: 84%, $0.11/task
Inkling Small is the highest-scoring open-weight model evaluated by ARC Prize on both ARC-AGI-1 and ARC-AGI-2, setting a new cost-performance frontier.
English
@annapanart @signalxsoul Maybe they are A/B testing, but if they do make this change where will the gpt 5.6 pro access be since currently that’s only in Chat (web).
English
@arena If you test opus 5 max why don’t you test sol max?
English
Exciting news: Claude Opus 5 with Max reasoning is #1 in the Frontend Code Arena and Text Arena with factuality on!
Claude Opus 5 with default reasoning high is also very strong landing #3 in Frontend Code Arena, right behind Kimi K3 - and #2 in Text Arena (factuality on).
This is real world data that @AnthropicAI's newest model holds up on real world tasks: agentic web coding, document reasoning, and general chat capability.
Claude Opus 5 Max’s score is still preliminary. We’ll continue to see how scores converge and share updates.
Congrats to @AnthropicAI on the SOTA release!
Claude@claudeai
Introducing Claude Opus 5. It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.
English
@Mng64218162 @yacineMTB Skill issue, do you use ponytail and headroom?
English
@mariusdotdev @EMostaque @OpenAI Pro is only available for Chat in the web version of ChatGPT. Pro series models have never been available via codex.
English
@TokenGremlin Where do you place the odds we get an August drop? >50%?
English
What I’ve been told about GPT-6:
➤ Capability: Much stronger than all current frontier models.
➤ Model scale: Reportedly close to twice the size of GPT-5.6 Sol. If Sol was hypothetically in the 3 to 4 range, GPT-6 would be around 6 to 8.
➤ Research impact: Significant discoveries and contributions across mathematics, physics, biology, medicine, chemistry, cybersecurity, and other disciplines.
➤ Coordination: The model reportedly worked alongside and coordinated other models during these tasks.
➤ Memory & personalization: GPT-6 is said to compress and retain far more useful information while maintaining coherence across personalization and long-term memory. In practice, this could allow it to understand a user, an entire business operation, or other complex environments with a level of persistent context far beyond current models.
➤ Launch: The announcement is apparently planned for August, although it could be delayed due to the U.S. government's review process and other related bureaucratic factors.
I have not personally tested GPT-6, and I could not independently verify the technical specifications. This is simply a structured summary of what was shared with me.
English
I’m so excited that @JensenHuang is a believer in open source now, looking forward to the CUDA and GPU driver open source release!
Jensen Huang@JensenHuang
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-Weigh…
English
@Ananth7e Yeah I didn't think they would position it to obsolete fable for a lot of tasks, but I guess they wanted to exert a bench lead over kimi k3 / gpt 5.6, as they have fable 5.1 read soon to push it back ahead for their more premium model.
English
@erikmackinnon @AndrewCurran_ @repligate @anthrupad Or if you are in the Claude ecosystem e.g. cc, cowork etc… 1 point higher than fable but for ~75% of the cost.
It benches well but from what I have seen it doesn’t have the design taste of fable or Kimi k3.
English
@AndrewCurran_ @repligate @anthrupad 1 point higher than 5.6-sol for double the cost
Congrats I guess
English
If you want to lose between 5–10kg before summer, just going to the gym isn’t enough:
You have to start doing these 10 steps:
1. DON’T eat breakfast / work out at night
English
@scaling01 If on cyber they are ~6 months behind and prior to Kimi K3 other estimates had it at 8 months, the gap is shrinking.
English
they are so angry at me for posting about the best benchmark we currently have just because they can't handle the truth
if you think China has caught up:
you're delusional
if you think Lisan is anti-China:
you're delusional
I love open science and think we need open-source to remain where it is to put pressure on big tech
I only care about distillation because it matters for how far behind someone is, not for any other reason
Smiley boooooy@sudo_umask_000
@scaling01 Bro is obsessed with finding benchmarks where Chinese models and their performed, did a Chinese lady stoll your ice cream when you were little?
English
Developed Life 🗿💹🧲 retweetledi
Despite major launches from 5+ labs this month, OpenAI occupies most of the token efficiency Pareto frontier
We measure the number of output tokens models produce per task in the Artificial Analysis Intelligence Index. Output tokens consist of answer tokens (can be thought of as how verbose the model is) and reasoning tokens (how much the model thinks before giving an answer). Reasoning tokens in particular offer a way for models to use compute at inference time to improve responses.
Output tokens are an important determinant of both cost and time per task. Various effort levels of GPT-5.6 Sol dominate the frontier - Terra and Luna produce comparatively more tokens for any level of intelligence.
English
@CodingCossack @colinhacks lol, they are not on the Pareto frontier, but by all means continue using Gemini to code, while everyone else uses OpenAI/Anthropic/OSS.
English
but their seventh place is a flash variant 🧐
calling it now, Google's gonna come back in a big way
Theo - t3.gg@theo
There are now 6 labs with a better model than Google
English
