Subnets / SN120
Affine SN120
Affine (subnet 120) is a Bittensor subnet held by 5,217 unique coldkeys. Its liquidity pool holds 78.2k τ, with the alpha token trading at 0.0490 τ. Roughly 0% of tracked holders are in profit — most bought in above today's price. Seller exhaustion is elevated: 78% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for Affine are available to SubnetStats subscribers.
ALPHA PRICE
0.0490 τ
IN TAO
HOLDERS
5,217
UNIQUE COLDKEYS
POOL DEPTH
78.2k τ
TAO IN POOL
7D MOMENTUM
-4.4%
ALPHA vs TAO
Project activity
A model doesn’t need to be the largest to matter. On Affine, models compete on how well their reasoning supports the next action across code, tool use and math.
A winner could put that reasoning to work in products, either directly or alongside a larger model. https://t.co/xMVTj
Affine submissions are now private, so miners can compete without exposing their weights to competitors.
Crowned models still go public. Losing checkpoints will be published later, so anyone can independently recompute every duel verdict. https://t.co/XUY3Ubkych
https://t.co/bQMyRNLQgQ
United against the divided.
Intelligence knows neither borders nor color. Uphold the torch with us to bring light where it is needed most. Open reasoning for humanity.
Join the thousand-year intelligence federation.
https://t.co/A9LUoBQvAd
One side effect of putting real money behind an AI evaluator is that the evaluator itself becomes an optimization target.
Since Affine switched to a new scoring system earlier this month, we’ve already seen a few strange equilibria. Empty or tiny cue-like thoughts were once comp
If your Hugging Face page is full of Qwen forks with names like 𝗳𝗶𝗻𝗮𝗹-𝘃𝟳-𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆-𝗴𝗼𝗼𝗱, you are probably closer to mining Affine than you think.
You already know the loop: change the data, recipe or objective, train another checkpoint, evaluate it, keep what works.
The differen
The recently released GLM-5.3 model is a good example of why post-training deserves its own competitive arena.
@Zai_org kept the same base model as GLM-5.2 and scaled post-training. More environments, more diverse tasks, more compute.
On Terminal-Bench 3.0, which tests agents
If you work on post-training or distillation, Affine is a live research competition with an incumbent you can dethrone.
GLM-4.5-Air-FP8 is frozen as the teacher. The corpus and scoring code are public, but your 1,300-turn evaluation slice is determined after you submit, so you c
We think the best Affine miner probably hasn't heard of Bittensor yet.
If you're an AI engineer or independent researcher who spends your time squeezing more capability out of open models, Affine gives that work a permissionless arena with real stakes. You don't need prior Bitte
Jacob Robert Steeves, co-founder of Bittensor and chief scientist at Affine, explaining the new reasoning-distillation incentive mechanism on Jason’s “This Week in Startups” podcast. https://t.co/X1Lp2KmLTW
Affine’s new incentive mechanism introduces a novel ‘reason-distillation’ strategy.
Traditional distillation runs into performance degradation as smaller models struggle to accurately reproduce the complex reasoning trajectories of larger ‘teacher’ models.
Both the teacher and
Qwen3.8-Max now available on LOGOS for closed beta users.
Join the waitlist at https://t.co/Qq6T0aaVl2
To grow beyond Kimi K3 at software engineering, Affine is not required to provide a model that surpasses K3 in every aspect.
We can leverage frontier models as our foundation and then push the individual capabilities that matter beyond what the base model is able to do on its ow
LOGOS enters closed beta.
Join the waitlist for open launch at
https://t.co/Qq6T0aanvu https://t.co/1qO4VF3bux
Periods of rapid technological advancement often intensify the same conflict: the expansion of human possibility against the impulse to restrict it in the name of safety.
As artificial intelligence undergoes its Cambrian explosion, a deeper question emerges: who is permitted to
Real environments anchor truth.
World models provide scale.
Verifiers apply judgement.
Learned environments are insufficient as substitutes for real ones.
Their value lies within being able to serve as the scale layer:
enabling cheaper rollouts, better curricula and faster it
Qwen3.6-35B-A3B now forms the new substrate beneath Affine's open arena.
By the very next day, the first champion had already been crowned. Following that, AFFINE-35B-III came along to break the frontier wide open:
Memory: 11.7 → 70.2
NavWorld: 33.8 → 48.3
SWE:
Emerging as the global frontier of model-reasoning has been a core vision of Affine since its inception. In pursuit of this vision, we've been working relentlessly to forge a repeatable training loop: a mechanism that uses benchmark driven post-training methodology to facilitate
Permissionless, Transparent, Decentralized.
Yes here at Bittensor Affine, you will find a real democratic place to train and get incentivized.
The 29th era of Affine Champion now.
While champions iterate fast, benchmark performance keeps raising the bar
AFFINE-XXIX vs. Qwen3-32B baseline:
• SWE-REBENCH +10.5
• SWE-MULTI +9.0
• HUMANEVAL +8.5
• MCP-AGENT +1.7
• BBH within tolerance https://t.co/UMWnAKx3a7
Honored to be invited by @Alibaba_Qwen and @alibaba_cloud to attend the very first global Qwen Conference in Singapore.
We’re truly grateful for the recognition of Affine’s work on agentic environments and post-training built on top of the Qwen ecosystem.
Affine will be joining
Great and productive time at today’s hackathon with experts from @Zai_org @alibaba_cloud , and the talented Chinese Bittensor community
We talked a lot about Affine’s agentic training environments and post-training infrastructure. One thing that feels more obvious now is that en
New Champion Affine-IV after a new week.
Significant improvements:
• TERMINAL: +12.3%
• MEMORY: +7.0%
• LIVEWEB: +3.1%
While all envs show no regression.
Four champions and each one stronger than the last.
Affine arena never closes and model evolution never stops. https://t.co
To improve mining experience, we built Affine GPU cluster with H200 on @TargonCompute.
Now inferencing orchestrated and optimzed on the Affine cluster.
• 2 x H200s per miner
• 8 slots for now - more to come soon
The arena now has its own compute. Expect full migaration soon. ht
Another week and another Champion.
Affine-III just dethroned Affine-II.
The SWE numbers tell the story:
SWE-Rebench: 0% → 5.3% → 10.5% → 14.0%
SWE-bench Multilingual: 2.3% → 10.3% → 10.7% → 16.0%
Three generations. Each one stronger than the last. https://t.co/PZubtyrPl7
SELLERS DRAINED
78%
28D SELLERS EMPTIED OUT
CAPITULATION
+0.15
BOUNCE COMPOSITE (z)
HEAT 7D
0.13×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
0%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling Affine — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →