🔴 Gemini 4 Neon xhigh is the best model in the world for 3D modeling when it comes to cost!
It's almost 70% cheaper than GPT 6.2 Astra and Fable 5.2, yet the results are even better, I still can't believe it
At this point, I'm convinced DeepMind has achieved RSI, Just look at the tests I ran and judge for yourself
Hydrogen was already really good with Blender, so I can only imagine what Neon will be capable of, I honestly think it could outperform every other model in Blender!
Gemini 4 Neon and Genie 4 are already here, and I'm honestly impressed
for example, this was built with Gemini 4 Hydrogen, I asked it to do it in 4K with Blender and it worked on it for almost 3 days, but the photorealism is on another level
It really feels like the limit is just your imagination because this model is incredible for 3D, Now I'm even more excited to test Gemini 4 Neon in 4K with Blender and see if it can match Opus 5.5 or maybe even beat it
👀 Gemini 4 Neon is now available on all paid plans, including Pro! I just tested it in Antigravity and it's so much faster than I expected, even better than Carbon at half the price. I'm seriously impressed, don't even know what to say!
🚨GTA 6 in 2030
I had GPT-6.2 Astra build a GTA 6 style game in one shot, but set in 2030 while keeping that Rockstar Vice City feel
Honestly, I'm really impressed. Astra is the best model I've used for 3D work, and I don't think even Fable 5.2 beats it right now
For this test I also used one Astra agent to audit every frame. I'm finishing a few final fixes now and then I'll release the public demo soon
Gemini 4 Neon, Krypton, Oxygen? 👀
I made this with Gemini 4 Hydrogen xhigh on my $20 plan, and 100% of it was done with Hydrogen. It took more than 2 days of rendering because in every cycle it reviewed each frame, and I also had a Hydrogen agent helping improve the modeling for the next round
It was made in Blender, but what really surprised me is how well the $20 plan handled two straight days of work without interruptions and still finished this piece
I really can't deny it anymore, this model has become one of my favorites
Gemini 4 Neon is coming in less than 8 hours! 👀
Pricing is $3 input and $15 output per 1M tokens, twice as expensive as Hydrogen, but it's expected to be on Carbon's level or even better. It'll also be multimodal like the rest of the Gemini models
I've personally tested Hydrogen with Blender on the $20 Pro plan and I'm still surprised by how well it performs Can't wait to see what Neon can do!
Google AI Ultra is a $0/month plan??
Right now a viral GitHub script claims it’ll hand you all of that free for 12 months
What’s not on the banner to run it, you type your Google password, your 2FA secret and your card into a stranger’s code.
not your 2FA code the secret. A code dies in 30 seconds. The secret generates every future code, foreve. Give it away and 2FA stops protecting you
and think about what that one login really opens: Gmail, Drive, Photos, YouTube your whole digital life, not just an AI app.
The “free” trial still needs a real card and autorenews at $99.99/mo. Google’s own team reportedly said it plainly: don’t register, it violates the Terms
a year of Ultra is worth a lot, Is it worth handing a stranger the keys to your entire Google account?
GPT 6.2 Sol and Astra are finally here, and for the first time OpenAI is bringing Ultrafast to the $200 Pro plan, starting with Sol! 👀
GPT 6.2 Astra beats Fable 5.2 by 5.5 points on Agents' Last Exam while costing about 95% less per task. It also takes first place on AA 3D with 1,318 Elo vs 1,271 for Fable 5.2
Astra's API pricing is down 20% too, and there's something even more interesting. It's the first publicly released OpenAI model to reach the High threshold for AI self improvement
Are you coming back to Codex or sticking with Fable 5.2?
GPT 6.2 Sol and Astra are finally here, and for the first time OpenAI is bringing Ultrafast to the $200 Pro plan, starting with Sol! 👀
GPT 6.2 Astra beats Fable 5.2 by 5.5 points on Agents' Last Exam while costing about 95% less per task. It also takes first place on AA 3D with 1,318 Elo vs 1,271 for Fable 5.2
Astra's API pricing is down 20% too, and there's something even more interesting. It's the first publicly released OpenAI model to reach the High threshold for AI self improvement
Are you coming back to Codex or sticking with Fable 5.2?
GPT 6.2 Astra Ultra is now available in Codex on all plans! 🔥
I can't believe how fast this model is, I think it could actually beat Fable 5.2 in benchmarks. Ultra mode with agents is incredibly efficient and barely uses any of my weekly limits on the $100 plan
I've only been testing it for a few minutes, but I already think this is the best model OpenAI has ever made. They should've called it GPT 6.5 Astra!
Fable 5.5 just spotted inside Devin Build? 👀
Two new stealth models have appeared, Silicon-2 with pricing at $10 and $50, which matches Fable exactly, and Roadrunner, whose pricing lines up with Gemini
I'm pretty convinced Silicon-2 is the next Fable model. Who else is charging those exact prices?
and with Anthropic's IPO reportedly just days away, could they already have their answer ready for GPT 6.2 Astra? The timing is getting really interesting!
Hy4 Flash High is now free on OpenCode with unlimited usage for two weeks! 🔥
I'm really liking this model so far, it's good and fast! Just look at the shadow GPT 6 Luna Max completely missed, Hy4 actually got it right and made the result look much more realistic
Hy4 took around 13 minutes for this test, compared to 15 minutes for Luna, with two weeks of free unlimited usage, I'm definitely going to use it as a subagent in Hermes Agent :)
Really liking this model!
Tencent just dropped Hy4 Flash, and the pricing is what surprised me the most 👀
In Tencent's blind test with 85 users and 259 real tasks, it came surprisingly close to Claude Haiku 5.5 in user satisfaction, scoring 2.79 vs 2.81
But here's the interesting part, each task cost just $0.04 compared to $0.71 for Haiku 5.5. That's around 18x cheaper! It was also the fastest model tested, taking just 6.7 minutes per task on average
And the specs are pretty impressive too. 224B parameters with only 12B active, a 1M token context window, native multimodal support and over 150 tokens per second , API pricing is just $0.07 per million input tokens and $0.28 for output
Compared to the larger Hy4, Tencent says it's 24.5x cheaper per task, with input costing 12x less and output 8.9x less, It's also free on WorkBuddy, with 50% off the API until November 30, and open weights are planned
of course, these are Tencent's own results, so I'm curious to see how it performs in independent benchmarks. But if these numbers hold up, Haiku 5.5 might have some serious competition, especially at this price
Qwen4-Max just scored 2.3x higher than Opus 5.5 on a legal AI benchmark 👀
21.3 vs 9.2 on HLAB-Vals, which tests AI agents on legal work. Every GPT 6 model scored below 7.2
But legal work isn't the only area where Qwen is doing well. It's also ranked #1 on:
• DeepSWE (coding agents): 75.7
• Agents' Last Exam (long tasks): 46.7, a 67% improvement over Qwen 3.8
• BioMysteryBench (bioinformatics): 85.0
It also took second place on AA-Briefcase, JobBench, APEX-Agents and AutomationBench
That's a top 2 finish in 9 out of 16 benchmarks against Opus 5.5, GPT 6.1 Sol, GPT 6 Astra and Kimi K3.1
Qwen 4 Max is officially available! 🔥
I ran this test with Qwen 4 Max Preview, and it went through 6 rounds of improvements on its own for my chip benchmarks. This is actually the first Qwen model that's really surprised me!
The whole thing took about a day, mostly because of rendering. It was visually checking every single frame and then going back to improve it again and again
And the best part? It only used around 50% of my $5 Qwen plan, which is pretty good considering the quality of the final result
I'd love to see what you guys can build with this model!
Qwen 4 Max is officially available! 🔥
I ran this test with Qwen 4 Max Preview, and it went through 6 rounds of improvements on its own for my chip benchmarks. This is actually the first Qwen model that's really surprised me!
The whole thing took about a day, mostly because of rendering. It was visually checking every single frame and then going back to improve it again and again
And the best part? It only used around 50% of my $5 Qwen plan, which is pretty good considering the quality of the final result
I'd love to see what you guys can build with this model!
GLM-5.5 could finally be dropping this week 👀
I've been looking at Zai's release history, and we're now 93 days since GLM-5.3 came out on August 14. That's the longest wait since GLM-4.5, beating the previous record of 83 days
Here's how the 2026 releases compare:
GLM-5: Feb 11
5.1: Apr 7 (55 days)
5.2: Jun 16 (70 days)
5.3: Aug 14 (59 days)
5.5: Still waiting...(93 days)
The average gap is around 61 days, meaning we could've expected the next model around October 14. We're now 32 days past that estimate and 23 days beyond the longest gap of 2026
I also checked the last 7 flagship launches. Three happened on a Tuesday, two on a Monday, one on a Wednesday and one on a Friday. Not a single Thursday or weekend release
My own estimate for this week is around 31%, and if I had to pick a day, I'd go with Tuesday, November 17