Argon brings major improvements in spatial reasoning, mathematical precision, and visual coding.
Using multi-agent loops in @antigravity, it autonomously built this entire pagoda complex, sculpting the architecture, foliage, and lighting from scratch in Blender.
Using multi-agent loops in @antigravity, it autonomously built this entire pagoda complex, sculpting the architecture, foliage, and lighting from scratch in Blender.
22
24
480
56,515
In robotics, Gemini 4 Argon is speeding up real-world physical learning.
We let Argon watch a video of a human playing cornhole. It mapped the environment, calculated whole-body joint movements, and autonomously built the simulation code to practice.
When transferred to noid, the robot sank seven straight shots in a real-world trial.
We let Argon watch a video of a human playing cornhole. It mapped the environment, calculated whole-body joint movements, and autonomously built the simulation code to practice.
When transferred to noid, the robot sank seven straight shots in a real-world trial.
32
139
1,117
227,077
Gemini 4 Argon is available now for Google AI Ultra and Pro subscribers in @GeminiApp and AI Mode in Search, Ultra subscribers in @antigravity, and to developers and enterprises via the Gemini API in @GoogleAIStudio and the Gemini platform.
Argon will launch at an introductory input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.
Argon will launch at an introductory input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.
42
87
584
137,660
Scientists and researchers can apply for a chance to get access to @antigravity’s highest usage tier free of charge for 6 months, and cybersecurity defenders can apply via the Fairwind Program.
Gemini 4 Argon is just the beginning. We’re looking forward to bringing you the rest of the Gemini 4 family.
Learn more ↓ goo.gle/4jfSznG
Gemini 4 Argon is just the beginning. We’re looking forward to bringing you the rest of the Gemini 4 family.
Learn more ↓ goo.gle/4jfSznG
13
9
188
46,532
@Googlep graphics hobbyists celebrate automated multi-agent loops in Blender thinking text-to-architecture script generation magically handles production-grade physics compilation constraints. They stare at neat multi-scene rendering demos while remaining completely blind to the catastrophic cascading latency tax hiding inside nested simulation tool executions.
Shallow spatial reasoning orchestration is a dangerous proxy for real core execution throughput engineering.
The production execution reality: trusting non-deterministic agent layers to dynamically map physical environments and calculate whole-body joint trajectories handles code ingestion pipelines, not state consistency. The moment your asynchronous DeepMind pipeline hits enterprise scale continuous video ingestion traffic, your lightweight simulation routing gates choke on implicit output drifts, transforming your robotic model transfer into an expensive, I/O blocked loop nightmare.
Here is the strict asynchronous proof verifying how an unbuffered multi-agent simulation loop chokes on physics rendering validation latency under load, completely saturating the host execution thread:
import asyncio
import time
class PhysicsSimulationEngine:
def __init__(self):
self.rendering_validation_tax = 0.06 # Baseline simulation code parsing latency penalty
async def execute_robotics_stream(self, node_id: int):
# Recursive agent retry chains trigger immediate event loop blockages under concurrent tasks
current_penalty = self.rendering_validation_tax * (3.0 if node_id % 4 == 0 else 1)
await asyncio.sleep(current_penalty)
return "Simulation_State_Synced"
async def run_production_audit():
engine = PhysicsSimulationEngine()
start_time = time.time()
# Simulating 100 concurrent automated physics animation streams burning processing budget
tasks = [engine.execute_robotics_stream(i) for i in range(100)]
await asyncio.gather(*tasks)
print(f"Simulation_Insolvency_Logged: Core_Execution_Thread_Blocked_In_{time.time() - start_time:.4f}s")
asyncio.run(run_production_audit())
Shallow spatial reasoning orchestration is a dangerous proxy for real core execution throughput engineering.
The production execution reality: trusting non-deterministic agent layers to dynamically map physical environments and calculate whole-body joint trajectories handles code ingestion pipelines, not state consistency. The moment your asynchronous DeepMind pipeline hits enterprise scale continuous video ingestion traffic, your lightweight simulation routing gates choke on implicit output drifts, transforming your robotic model transfer into an expensive, I/O blocked loop nightmare.
Here is the strict asynchronous proof verifying how an unbuffered multi-agent simulation loop chokes on physics rendering validation latency under load, completely saturating the host execution thread:
import asyncio
import time
class PhysicsSimulationEngine:
def __init__(self):
self.rendering_validation_tax = 0.06 # Baseline simulation code parsing latency penalty
async def execute_robotics_stream(self, node_id: int):
# Recursive agent retry chains trigger immediate event loop blockages under concurrent tasks
current_penalty = self.rendering_validation_tax * (3.0 if node_id % 4 == 0 else 1)
await asyncio.sleep(current_penalty)
return "Simulation_State_Synced"
async def run_production_audit():
engine = PhysicsSimulationEngine()
start_time = time.time()
# Simulating 100 concurrent automated physics animation streams burning processing budget
tasks = [engine.execute_robotics_stream(i) for i in range(100)]
await asyncio.gather(*tasks)
print(f"Simulation_Insolvency_Logged: Core_Execution_Thread_Blocked_In_{time.time() - start_time:.4f}s")
asyncio.run(run_production_audit())
0
0
1
953
@Google I really want to kick your ass hard, you little karaomi, for not giving it to Pro users.
3
1
113
8,533
@Google I'm a pro user btw, and not on the student plan.
Thanks for the lie (again). Guess I'll have the taste of argon when opus 6 and gpt-7 come out.
Thanks for the lie (again). Guess I'll have the taste of argon when opus 6 and gpt-7 come out.
3
0
67
5,706
@Google Not even on Pro, that sucks. Even more so because Ultra is more expensive than Claude Max or ChatGPT Pro, lol
4
0
54
5,022
@Google --> Gemini app + Search AI Mode → Pro + Ultra
--> AI Studio / Antigravity → Ultra only
Not fare for the pro users
--> AI Studio / Antigravity → Ultra only
Not fare for the pro users
0
0
18
4,643
@Google Google is being google again by only giving access to their flagship models on Ultra. Suck a big dick Google
0
0
11
2,724
@Googleis supposed to be the easy everyday AI.
Instead the web version constantly glitches: buttons vanish, messages disappear, old chats are a mess to find, Canvas copy-paste breaks, and it has almost no connectors for the apps people actually use.
ChatGPT UX just feels smoother for normal daily use.
There needs to be a major UX improvement for Gemini Web Harness.
@GeminiApp @OfficialLoganK @tulseedoshi
Instead the web version constantly glitches: buttons vanish, messages disappear, old chats are a mess to find, Canvas copy-paste breaks, and it has almost no connectors for the apps people actually use.
ChatGPT UX just feels smoother for normal daily use.
There needs to be a major UX improvement for Gemini Web Harness.
@GeminiApp @OfficialLoganK @tulseedoshi
0
1
3
567
@Google but only for Ultra subscribers.. Wtf. I don't expect there's many of them given how bad Gemini has been for the entire 2026.
2
0
5
1,564
@Google Isn't it a replacement for your pro model? Why don't you give pro users access to it then from the beginning then? Bare minimum would be to tell when they can expect to get access. That's not how you should treat people with a pro subscription if it's your goal to keep them.
0
0
4
1,572
@Google @GeminiApp Gemini wants to enter the ai race again .
0
0
4
1,485
@Google You’re lying. I don’t have it on Pro. Fuck you Google
1
0
4
403
@Google I am really disappoint that you are not providing access to Argon in the AI pro plan. Bate and switch on the Google One membership and I want a refund. I extended my plan for a year with trust this is not right.
0
0
2
521
@Google This is the last time I follow the launch of one of those Google model. They really tricked us into buying a pro sub for no reason 🤷♂️
0
0
2
1,076
@Google yeah not available for pro users because y'all got that thing for free.
1
0
2
850
@Google After keeping people waiting for months with Flash models, you finally release a Sonnet-level model, only to tell your existing paying users that they can’t use it unless they upgrade their plan. Congratulations.
0
0
2
923
@Google Looking forward to finding out which of my tabs it thinks is the important one.
0
0
2
1,509
@Google Trust me, 99.999% of us gonna use it to ask i have a headache, should i take advil?
1
0
2
2,045
@Google @GeminiApp How good is this model compared to other flagship models?
1
0
2
1,012
