Quasar is one of the cleaner recent examples of a Bittensor subnet giving the market an object to inspect.
The object is public. Readers can inspect a Hugging Face model card, an 18B class MoE architecture, a 2B active parameter path, an experimental 5 million token context configuration and caveats written directly into the release.
This already gives readers more to inspect than a vague claim that a subnet is doing decentralized AI.
What Quasar released
The public model is silx-ai/Quasar-Preview on Hugging Face.
The model card describes it as the first public checkpoint in the Quasar Foundation Model series. The tags and README describe an 18B class sparse Mixture of Experts model with roughly 2B active parameters, hybrid recurrent and attention layers, and an experimental 5 million token context configuration.
The important word is experimental.
The model card says the 5 million token path has received less than 1B tokens of long context extension training so far. The configuration exists, but the release should not be marketed as mature 5 million token reasoning.
That caveat makes the release more credible, not weaker.
Why SN24 matters now
On June 23, the Quasar account said it was launching a 10T token incentive mechanism. The first stated target was 5T tokens.
The live Bittensor angle is the incentive process around SN24.
Quasar is trying to turn model improvement into a subnet process. The public thesis is that SN24 can help improve a model through decentralized training, distillation, evaluation and iteration, rather than rewarding activity that never touches the model.
The problem is hard.
It gives readers a practical checklist. Open the Hugging Face page, read the model card, check the checkpoint date, look for miner submission data and watch whether training claims become measurable model improvement over time.
The emission confusion
Quasar also had to explain an emission issue after the broader Bittensor cleanup.
The team said SN24’s burn behavior could look like 100 percent burn when there was no winner, then return to 0 percent when a winner received emissions. The team said winners had been paid and that it changed the burn behavior to avoid confusion.
The episode matters because it shows the new standard.
Subnets now need to make their mechanisms readable from the outside. If on chain behavior looks inactive or extractive, even a real project can get caught in the cleanup logic.
Quasar has a strong product object. It still needs to keep the incentive surface easy to verify.
What to watch next
The first watch item is whether the 10T token training mechanism produces public artifacts. The second is whether SN24 can show model progress in a way readers can verify. Benchmarks are useful only when the harness, prompt format, checkpoint and comparison method are clear.
The third watch item is whether miners can participate without the process becoming opaque to outsiders. The fourth is whether the market values a research heavy subnet differently from an inference or compute subnet.
Quasar does not need to be perfect today. It needs to make the next checkpoint more convincing than the current one.
The Tao Outsider read
Quasar Preview is interesting because it gives Bittensor a model card instead of only a narrative.
The 18B MoE headline will get attention. The real test is slower.
Can SN24 turn decentralized training into measurable improvement?
Can miners compete on something that advances the model?
Can the team keep the public evidence readable enough that the market understands what is happening?
If the answer improves over the next checkpoints, Quasar becomes a serious long context story inside Bittensor.
For now, the correct read is cautious interest. Quasar has a public checkpoint, a real technical surface, an ambitious incentive mechanism and clear caveats. That is enough to keep watching.
Sources
Hugging Face: silx-ai/Quasar-Preview
Quasar X update: 10T token incentive mechanism
Quasar X update: emission block explanation
TAO Media: Quasar and its 10 trillion token run is trending on Hugging Face
Source trail
What this article was checked against
Tao Outsider preserves the primary source path whenever possible. Links below are extracted from the article source section for faster verification.
- silx-ai/Quasar-Preview huggingface.co
- 10T token incentive mechanism x.com
- emission block explanation x.com
- Quasar and its 10 trillion token run is trending on Hugging Face tao.media
- Author
- Iris Vale
- Reviewed by
- Tao Outsider
- Scope
- News report
Was this article useful?
One tap feedback helps us improve each post.
