CTC Round 1 Live
Chain of Thought Compression Β· Upload Window Open New semantic scoring β€” precise answers score 0.75–1.0 Β· Dataset: real CoT SWE coding sessions
--days
:
--hrs
:
--min
:
--sec
Prepare Now β†’
⚑ Bittensor SN114 · 6 Rounds Completed · CTC Round 1 LIVE Now

The AI Compression Race

Compress smarter. Score higher. Earn TAO. Track all rounds, every miner, live β€” now with Chain of Thought semantic compression.

⚠️ Unable to reach SOMA API. Data may be stale.
CTC R1
Current Round
β€”
Miners
β€”
Qualified
β€”
Top Score
β€”
Avg Qualified
β€”
All-Time Unique
πŸ“Š All Rounds
Click any round to view its leaderboard Β· CTC Round 1 is the new Chain of Thought competition format
Top Score Per Round
Old compression
CTC format
πŸ† Leaderboard
Loading…
# Hotkey Screener Score β–Ό Final Score Scores Status Last Submit
Loading…
⛓️ SN114 On-Chain
Live on-chain data for Bittensor Subnet 114 β€” SOMA
114
Subnet ID (Netuid)
~256
Registered Miners
7
Competition Rounds
β€”
Unique Competitors
πŸ”— View SN114 Live on TaoStats β†’ 🌐 SOMA Platform β†’
πŸ“– What is SN114 SOMA?
From text compression to Chain of Thought β€” the evolving AI compression competition

"Efficient compression is foundational to AI."


SN114 SOMA uses Bittensor's economic incentive layer to drive a global competition: who can build the best AI compression algorithm? Real TAO for real performance β€” verified on-chain.


Rounds 1–6 tested text compression at four ratio targets (0.2 / 0.4 / 0.6 / 0.8). Top score: 79.82% in R5.


CTC Round 1 (May 4–7) brings a new format: Chain of Thought QA compression with semantic scoring. Precise answers with key qualifiers (dates, counts, conditions, negations) score 0.75–1.0. Vague or echo answers score 0.0.

🧠
Chain of Thought Compression (CTC)
New format: compress CoT reasoning traces from real SWE coding agent sessions. Semantic scoring β€” meaning preserved = high score.
🎯
Semantic Bucket Scoring
Fixed scores: 0.0 / 0.5 / 0.75 / 1.0 β€” no partial credit. Precise answers with key qualifiers (dates, counts, negations) unlock 0.75–1.0.
βœ…
Two-Stage Evaluation
Screener test β†’ qualify β†’ final score β†’ emissions. Pass the threshold to earn real TAO. LLM_MAX_TOKENS upgraded to 4096 for CTC.
πŸ“ˆ
Rising Competition
R1 threshold ~50% β†’ R5/R6 peaked ~79%+ β†’ CTC resets the field with new semantic metrics. First-mover advantage is real.
CTC Round 1 β€” Score Buckets
1.0
Perfect
Precise answer with ALL key qualifiers (dates, counts, conditions, negations)
0.75
Good
Mostly correct with most important qualifiers β€” minor omissions acceptable
0.5
Partial
Correct direction but missing key details or conditions
0.0
Failed
Vague, echo, wrong, or missing answer β€” no credit
πŸš€ How to Join
Start competing and earn TAO emissions
1
Register on SN114
Burn ~0.9 TAO to register a hotkey on subnet 114.
btcli subnet register --netuid 114 --wallet.name WALLET --wallet.hotkey HOTKEY
2
Build your CTC agent
Python agent that compresses CoT reasoning traces with precise, qualifier-rich answers. LLM_MAX_TOKENS is now 4096. Bucket scores: 0.0 / 0.5 / 0.75 / 1.0.
3
Upload during window
Submit your solution file via the SOMA upload tool. Upload window: check thesoma.ai/dashboard for current round.
python3 upload_miner.py --platform_url https://platform.thesoma.ai --wallet_name WALLET --hotkey_name HOTKEY --solution_file miner.py
4
Pass screener & earn TAO
Hit the qualification threshold β†’ earn emissions. Iterate to climb the leaderboard.
πŸ“‹ Competition Rules
CTC Round 1 format
SubnetSN114 β€” SOMA
FormatCTC β€” Chain of Thought
Score buckets0.0 Β· 0.5 Β· 0.75 Β· 1.0
LLM_MAX_TOKENS4096 (was 1024)
Upload windowSee thesoma.ai/dashboard
Top score (R6)79.19%
Qualify thresholdloading…
Platformthesoma.ai
πŸ’‘ CTC Strategy: Scores are semantic buckets β€” either 0.0, 0.5, 0.75, or 1.0. Maximize 0.75–1.0 scores by providing precise answers with dates, counts, and exact conditions. Vague or echo answers = 0.0.
Live refresh in 60s