Jev is not open-source and only available via API.
Here's an open version called Nimble which performs just as well.
New open source music generator, YuE2
Even beats Mureka v9 and Minimax 3 on various benchmarks
Another realtime method for Minimax H3.
They used 8x B300's though ๐ Curious how this performs on consumer GPUs.
While benchmarks are being crushed by the latest AI models, it's funny how the frog test still holds up well.
Might be the hardest test left.
GPT 6 Astra could not find the frog ๐ญ๐ญ๐ญ
Y'know, Google had an Astra too. What happened to it???
I asked GPT 6 to compose a piano solo, but that's not all. It had to play it LIVE on an online piano, by pressing the keys one by one.
Full review:
Finally, a speedup method for Minimax H3 reference to video!
ComfyUI-Ref2VA-VSA adds video sparse attention to Minimax, generating 5-second clips in ~72s on an RTX 4090.
It keeps reference tokens dense while pruning video attentionโ2.24ร faster than VDN-H3 at ~13.5GB VRAM.
Show more
GPT 6 Astra composing an EDM track from scratch in my DAW
GPT 6 Astra crushes 3D game design. Here's a sneak peek of one of my tests.
If you're vibe coding games or 3D assets with GPT 6 Astra, by default it'll still output pretty basic, sloppy-looking visuals...unless you use this prompt hack.
You need to include a variant of the following in your prompt.
Get it to spin up a separate critic agent, which takes screenshots of your generation at different angles and gives it a score from 0 to 10 (define what this means). Then, get it to keep looping until it achieves a score of >8.
More details in my upcoming review video.
Show more
Add this new adapter to Minimax H3 for much better prompt adherence.
If you're vibe coding games or 3D assets with GPT 6 Astra, by default it'll still output pretty basic, sloppy-looking visuals...unless you use this prompt hack.
You need to include a variant of the following in your prompt.
Get it to spin up a separate critic agent, which takes screenshots of your generation at different angles and gives it a score from 0 to 10 (define what this means). Then, get it to keep looping until it achieves a score of >8.
More details in my upcoming review video.
Show more
GPT 6 Astra spent 3 hours making this, and is still going. Let's see how far I can push it. Full review coming soon
GPT 6 Astra usage stats for your reference.
Unlike Fable 5.1 by Scamthropic, running GPT 6 Astra (Extra High) for an hour only uses around 4% of my weekly usage limit. I'm on the Pro 5x plan.
GPT 6 Astra could not find the frog ๐ญ๐ญ๐ญ
If you're waiting for GPT 6 Astra, check again! You should have access now
SolarWM turns Minimax H3 into realtime interactive worlds. It can keep generating for over an hour.
Highlights from GPT 6 Astra 117-page system card:
First model to hit OpenAIโs Critical cyber threshold: can find unknown flaws and write exploits on hardened systems with little human guidance.
Protections scaled up: isolation, encrypted checkpoints, full-trajectory/CoT monitoring, blocking alignment evals, paused non-compliant internal work.
Far more jailbreak-resistant than GPT-5.6 Sol, including long chats; high-risk users can get stricter refusals (~91.5% vs ~59% on cyber jailbreaks).
Better aligned: ~half as many high-severity misalignment flags as Sol on 54k internal Codex tasks.
Misalignment monitoring now runs on all tool-using external Astra inference and can stop risky agent work.
Harder to monitor: more CoT control, sandbagging, and some monitor evasion when told to hide; no steganographic CoT seen.
Stronger against prompt injection and less likely to take destructive workplace/browser actions.
Safer answers overall, including for teens (big gains on eating-disorder and age-restricted content).
Better HealthBench and mental-health results; fewer user-flagged hallucinations.
100% on ExploitBench even at lowest reasoning; other cyber benches plus external review support the Critical rating.
Bio/chem still treated as High, not Critical; research access is gated.
Rollout is phased and identity-gated (Trusted Access, extra security, high-risk jurisdiction limits).
Show more
GPT 6 Astra numbers are insane:
98.6% on ARC AGI 3, 100% on Exploit Bench
Can't wait to use it
VDN Minimax H3 for near lossless speedup now works in ComfyUI.
Another HUGE speedup method for Minimax H3.
VideoDeltaNet offers up to 90x speedup with "near lossless quality". Higher quality and better instruction following than FastH3.
Show more
Another HUGE speedup method for Minimax H3.
VideoDeltaNet offers up to 90x speedup with "near lossless quality". Higher quality and better instruction following than FastH3.
Show more