ʟᴇɢɪᴛ
@legit_api
Followers
9K
Following
31K
Media
876
Statuses
3K
api guy with little-to-no wisdom, 0.5x engineer
Joined October 2018
Made by Veo 3. "A squirrel attempts to launch a homemade rocket into space"
14
32
388
this is my mini benchmark for AGI. if it can see and name them, it's agi. currently no model can see them 👀
331
273
9K
this is so unbelievable true. I’m incredibly grateful that OpenAI shipped gpt-3. the singularity is within reach 😌
29
69
1K
- Gemini 2.5 Flash.- Veo 2 editing capabilities.- Updated Imagen 3 + inpainting.- Lyria (text-to-music) now in preview.- TPU v7 (Ironwood) soon™.- Chirp 3 HD Voices + voice cloning
21
57
649
Llama 4 Omni is releasing very soon 👀. my system has detected a new official page for the upcoming models
23
52
563
10 MILLION TOKEN CONTEXT WINDOW. holy cow Llama 4!
Llama 4 Omni is preparing for launch. these new pages were added today. official release any day this month 🤷♂️
32
32
557
@Willrandship I do not as it changes the image still. but that's a clever way to work around this, nicely done.
4
0
538
you can prompt it how ever you like as long as it's reasonable. I simply hope to see more work put into visual reasoning for 2025.
8
2
512
Gemini 2.5 Pro pricing now official. - $1.25/1M input for <200K tokens.- $10/1M output for <200K tokens. - $2.50/1M input for >200K tokens.- $15/1M output for >200K tokens
17
33
471
an updated 2.0 flash thinking model snapshot is releasing in a few days. I do have the date but I'll let them announce it on their own terms 🙂
17
41
434
Deep Research agent first impressions - relatively big deal, significant time savings and task delegation that actually works!. BUT not a fan of how openai is using the Pro tier as the new testing ground for new features - this was the entire original premise of the PLUS tier 😞
today we launch deep research, our next agent. this is like a superpower; experts on demand!. it can go use the internet, do complex research and reasoning, and give you back a report. it is really good, and can do tasks that would take hours/days and cost hundreds of dollars.
26
18
459
A conversation with OpenAI on what to expect in the next 12 months. - today's systems are laughably bad.- ChatGPT not long term engagement model.- models capable of complex "work".- like a great team mate working with u.- shift towards verbal interfaces & beyond, Multimodality
28
79
433
here's an exclusive new udio AI music generation 🫡. source: anon. anon is always the source.
44
65
403
Grok 3 launching very soon. I managed to sneak my way in, send prompts fast before i get cut off!
15
17
395
nightwhisper is an insane model at coding. added today to LM Arena and WebDev Arena 👀. "Build me a working prototype of the X Platform"
21
28
379
Llama 4 Omni is preparing for launch. these new pages were added today. official release any day this month 🤷♂️
23
19
360
Imagen 3.5 and Veo 3 coming soon!. these *might* be intended for I/O 👀
24
27
351
Veo 2 is likely going to ship inside Gemini soon. these strings which were previously in beta have now been added in the latest stable release 16.11.37:
8
30
343
this is so true, and I'm here for it 🍿. every AI lab is cooking, closed & open
7
17
325
Deep Research has been upgraded to Gemini 2.0 Flash Thinking?. big if true
25
10
329
New Gemini 2.5 Pro Preview model is claybrook. for me, Nightwhisper was consistently better. I hope they release that model at some point 🗿
14
18
326
there's a chance that the next Gemini model to be released is 2.5 Flash. hopefully there's more after this
15
8
305
preparations for OpenAI o1 confirmed. the model we got a glimpse of was o1 ✨
6
23
295
the o1 party is over 😢
14
5
295
we're getting multiple Gemini 2.0 models. some are already up for testing on LM Arena including 2.0 Flash (stable) + 2.0 Pro (unclear if it's exp or stable). and there's another thinking model in testing that thinks longer than the latest one to produce better results 👀.
12
13
296
AI Studio is ready for the launch of the new Gemini 2.0 models. "Gemini 2" category was added yesterday in the model selector so the stable versions will be listed there as soon as they release. the question is WHEN? and to that idk.
8
17
291
Grok 3 and Grok 3 mini MIGHT have entered LM Arena ahead of their upcoming release. based on some tests / comparisons to the Grok 3 model I tested a few weeks ago, it outputs similar code and avoids name dropping itself or xAI. i think Grok 3 is "chocolate" and Mini is "kiwi"
24
16
294
GPT-4.5 blog post is prepared 👀. awaiting release possibly today
10
15
286
o4-mini and o3 should be releasing this week 👀. these are ofcourse reasoning models, fun!
12
13
290
Coding agents with an entry point in ChatGPT coming. It will not be inside ChatGPT hence the added string: "Back to ChatGPT". - "Asynchronous coding agents".- "Write code, fix bugs, draft PRs"
15
36
283
GPT-5 has been spotted by @Jaicraft39 on the official copilot pricing page from Microsoft 👀. surely this can't be a typo right. cc @apples_jimmy @sama
26
39
267
it's time 😮💨. - Gemini 2.0 Pro Exp (experimental).- Gemini 2.0 Flash (stable). we should also be getting a tech report with it!
we're getting multiple Gemini 2.0 models. some are already up for testing on LM Arena including 2.0 Flash (stable) + 2.0 Pro (unclear if it's exp or stable). and there's another thinking model in testing that thinks longer than the latest one to produce better results 👀.
5
9
267
grandma is trapped in this cage until. GPT-5 | Gemini 2 | Grok 3 | Llama 4 😇
15
16
250
oai will make an announcement later today for the release of the o1 model 👽. info provided by @BadArtist154
5
22
259
Google is currently testing an experimental Imagen model 👀. unclear if this will be Imagen 4 or an updated imagen 3. btw their big I/O event is next month
15
19
258
nightwhisper has left the Arena 👀. the insanely capable coding model ^. Veo 2 is being prepared for AI Studio and the Gemini API
11
11
253
In an OpenAI Event at University of Tokyo:. Sam discusses future direction of development; "GPT-5 and GPT-6, [. ], will utilize reinforcement learning and will be like discovering new science, such as new algorithms, physics, and biology.".
Q. 今後の開発の方向性は？.A. GPT-3やGPT-4は、Pre-Trainingのパラダイム。今後開発していくGPT-5、GPT-6では強化学習も活用し新しいアルゴリズム、物理学、生物学など、新しい科学を発見していくようなものになるだろう。.
7
25
244
GPT-4.5 (Orion) seems really good at drawing in svg, in fact the best by a long shot. if this is any indication of the rest of its capabilities, pretraining + postraining may still show big gains. confidence in it being 4.5? 100% 😊.
13
10
248
GPT-4.5 research preview releasing for Pro users. wonder if other plans will get it too 🥺.
10
12
244
Claude Research advanced mode is now available. it can research for up to 45 minutes!
Today we're announcing Integrations, a new way to connect your apps and tools to Claude. We're also expanding Claude's Research capabilities with an advanced mode that searches the web, your Google Workspace, and now your Integrations too.
7
19
246
Gemini 2.5 Pro / Flash auto caching!. - lowered min context: 1k / 2k (flash/pro).- automatic and effortless caching.- SAME 75% TOKENS DISCOUNT!!
7
20
244
in collaboration with Kol (@koltregaskes) we've launched a Notion page to track upcoming model updates. ranging from GPT-5 (Orion) & o3, to others like Gemini 2.0 Pro, Claude 3.5 Opus, Claude 4, chatbot arena models
10
10
241
new gpt-4o model updated with June 2024 knowledge cutoff is back 👀. it might be on rollout again so not everyone will have it but go check.
12
7
237
Veo 3 is rolling out to Pro tier!. the shipping continues in Gemini web
53
30
391
haha o3 didn't spend a single thinking token on my pleasantries. must be token efficient 👀
7
9
236
Microsoft has a purchase agreement with Helion for electricity through commercially viable fusion energy by 2028 ⚡. would make alot of sense for the Stargate ai supercomputer for openai to be powered by this as it's also expected to come online as soon as year 2028
Singularity is nearest? .MICROSOFT AND OPENAI PLOT $100 BILLION STARGATE AI SUPERCOMPUTER .MICROSOFT EXECUTIVES ARE LOOKING TO LAUNCH STARGATE AS SOON AS 2028.OPENAI’S NEXT MAJOR AI UPGRADE IS EXPECTED TO LAND BY EARLY NEXT YEAR
16
33
227
@Oli82817545 a theoretical agi would be well informed of pareidolia and be able to identify to a high degree if the image arrangement is an example of such. a general pattern I see as models scale in size is that they become more perceptive but not at the same rate for image modality.
5
1
235
HUGE - o3 and o4-mini is getting support for scheduled tasks. this will let people automate tedious work, the shipping never ends!
14
12
233
gemini 2.0 flash now available on cursor pro. add it to models list and disable your own api key
9
15
223
Google I/O starts in 10 days! 👀. NO FURTHER SPOILERS ALLOWED. hold your excitement for the event!
18
13
228
specter - a new model in the arena, feels like a snapshot of 2.0 Pro Thinking. very capable and warm at chat, strong + slow reasoning, very perceptive of the ongoing context. 2.0 Flash Thinking stable tomorrow unless delays
15
10
219
"GPT-5 is our next foundational model". "everything our models can currently do better and with less model switching". Improved Operator coming soon also. They also want to bring all the tools together so it feels like one thing
Jerry Tworek, OpenAI VP of Research talking about combining multiple products into GPT-5. This is from the current reddit AMA.
8
18
219
it's all coming together. - stargazer seemed like 2.5 Flash to me.- I didn't get to try lunarcall. both left the arenas 2 hours ago!
gemini-2.5-flash-preview-04-09 + thinking_config/thinking_budget have been added to Google Gen AI Python SDK
20
8
218
anonymous-chatbot is back in the arena. this is usually reserved for GPT-4o model updates inside ChatGPT 👀
5
11
210
Sam confirms OpenAI wants to bring the O series and GPT series together. now it makes sense why he's talking about GPT-5 and GPT-6 again. we are going to get an "everything" model that knows when to reason, when to respond fast, when to carry out research, when to use voice, etc.
Sam Altman says the leap from GPT-4 to GPT-5 will be as big as that of GPT-3 to 4 and the plan is to integrate the GPT and o series of models into one model that can do everything
6
15
205
I'm teaming up with Bedros, M1, Kol, . and TestingCatalog to bring faster . and more accurate AI insights
21
12
207
@K3vn_C understandable but modern llms should know about Family Guy and approximate cultural aspects. GPT models can sometimes fall for the same illusions as humans so this gives me hope it is achievable. natively multimodal models need more work as we increase reliance on them.
5
0
199
Grok 3 launching on Feb 17th (Monday). kiwi has left the Arena & I *think* chocolate might be gone too. Elon mentioned he'll be offline for final tweaks ahead of a demo & release.
7
11
197
g(o)(o)d m(o)rning 🌅
good morning.
7
7
192
for the record, this is the best news I've heard since the O series of models!. if GPT-5 is on the way, then we are almost certainly going to get another Claude Opus + more from others 👀. even 1 of these alone could be a godsend so I pray that everyone involved can pull it off
@kimmonismus @flowersslop 1 and 2, figuring that out but i think you'll be happy. 3, i would love for us to be able to merge the GPT series and the o series in 2025! let's see.
6
8
191
GPT-4.1 doing better than GPT-4.5, o3-mini-high, o1. 1. this is awesome 🔥.2. did not think naming could get stranger
8
8
198
Gemini 2.0 Flash - native image output 📸. now available on 2.0 Flash Experimental model. you may need to regenerate until it stops erroring
9
8
199
the lightweight version of Deep Research in ChatGPT seems to be getting prepared for release. Free users and above will get access to this!
10
10
198
Grok 3 update - I can now confirm that "kiwi" is a Grok model so chances are "chocolate" is also one. these new Grok models are much better than their last model in terms of code and creativity. "chocolate" feels like full grok 3, "kiwi" feels like mini or quantized version.
Grok 3 and Grok 3 mini MIGHT have entered LM Arena ahead of their upcoming release. based on some tests / comparisons to the Grok 3 model I tested a few weeks ago, it outputs similar code and avoids name dropping itself or xAI. i think Grok 3 is "chocolate" and Mini is "kiwi"
17
8
192
We might get Imagen 4 & 4 Ultra. and skip 3.5 entirely. let's see, 3.5 might be for internal use
7
11
197
Optimus Alpha, added today on OpenRouter. - is super fast, currently faster than Quasar Alpha.- both claim to be based on GPT-4. both now claim June 2024 cutoff like gpt-4o
7
11
194
nightwhisper is still better than dragontail. dragontail frequently produced errors for me or worse formatting when it produced something usable. dragontail might be better than 2.5 pro but nightwhisper was consistently a level above with exceptional formatting + no errors!.
nightwhisper is an insane model at coding. added today to LM Arena and WebDev Arena 👀. "Build me a working prototype of the X Platform"
15
10
192
@JaimeOrtega a pretty useful thing to learn in the context of theory of mind + if we want truly superhuman perception.
1
1
188
Grok 3 - if reasoning / thinking is enabled by default, then I can see why xAI is super hyped about what they've made. I'd also start to believe the "smartest AI on Earth" claim by Elon 🧐
still have a few surprises left :^).
11
6
183
calling all ai leakers - blink twice if Llama 3 is still dropping today
14
16
176
Gemini 2.0 Flash & Gemini 2.0 Pro Exp confirmed for tomorrow? 👀. Is the wait finally over? or will they edit this and call it a typo again 😑
11
10
179