OSC Open Superconductor Challenge
๐งฒ
32
Estimate and submit dโwave pairing scores for superconductors
seawolf PRO
seawolf2357
AI & ML interests
None yet
Recent Activity
liked a Space about 15 hours ago
FINAL-Bench/OSC-Leaderboard reacted to SeaWolf-AI's post with โค๏ธ 1 day ago
๐งฌ Darwin-180B-RSI โ an AI that learns from itself and knows when it's right
๐ https://huggingface.co/FINAL-Bench/Darwin-180B-RSI
๐งฌ Darwin โ crossbreed and evolve the parent
Darwin diagnoses strong parent models like an MRI, inherits only their best parts, and evolves the weak spots โ producing a child stronger than its parents.
Father model: Qwen3.8-Flash-Next (180B MoE).
๐ง Rewired paths
๐น 12 full-attention layers ยท ๐น 36 linear-attention layers ยท ๐น 48 shared-expert layers โ precision-strengthened
๐ 512 routed experts ยท router ยท vision encoder โ untouched
โ Only 0.02% of the weights changed.
๐ RSI ร ๐๏ธ ZTC
RSI (recursive self-improvement): solve โ verify against real answers โ learn only the correct reasoning โ repeat.
ZTC (Zero-Token Confidence): reads the model's internal state once, before answering, and returns the probability the answer is right โ zero extra tokens. Returns answer + confidence as JSON.
{"answer": "...", "confidence": 0.97, "truncated": false}
โจ Synergy: ZTC finds where the model wavers โ RSI learns exactly there โ confidence gets sharper. Low confidence = stop, so agents don't act on wrong answers.
โก Same accuracy, 11% shorter reasoning โ faster and cheaper.
๐ https://arxiv.org/abs/2605.14386
๐ค https://huggingface.co/FINAL-Bench/Darwin-180B-RSI
๐๏ธ https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
๐ The result โ #1 on five Hugging Face official leaderboards
๐ฅ AIME 2026 100% (first perfect score on the board)
๐ฅ HMMT Feb 2026 100% (first perfect score on the board)
๐ฅ GPQA Diamond 94.44%
๐ฅ MMLU-Pro 88.12%
๐ฅ MMMU-Pro 79.48%
๐ 131K-token thinking budget ยท bf16 ยท samples per benchmark listed on the model card. ๐
#Darwin #RSI #ZTC #AIME #HMMT #GPQA #MMLUPro #MMMUPro #OpenSource reacted to SeaWolf-AI's post with ๐ 1 day ago
๐งฌ Darwin-180B-RSI โ an AI that learns from itself and knows when it's right
๐ https://huggingface.co/FINAL-Bench/Darwin-180B-RSI
๐งฌ Darwin โ crossbreed and evolve the parent
Darwin diagnoses strong parent models like an MRI, inherits only their best parts, and evolves the weak spots โ producing a child stronger than its parents.
Father model: Qwen3.8-Flash-Next (180B MoE).
๐ง Rewired paths
๐น 12 full-attention layers ยท ๐น 36 linear-attention layers ยท ๐น 48 shared-expert layers โ precision-strengthened
๐ 512 routed experts ยท router ยท vision encoder โ untouched
โ Only 0.02% of the weights changed.
๐ RSI ร ๐๏ธ ZTC
RSI (recursive self-improvement): solve โ verify against real answers โ learn only the correct reasoning โ repeat.
ZTC (Zero-Token Confidence): reads the model's internal state once, before answering, and returns the probability the answer is right โ zero extra tokens. Returns answer + confidence as JSON.
{"answer": "...", "confidence": 0.97, "truncated": false}
โจ Synergy: ZTC finds where the model wavers โ RSI learns exactly there โ confidence gets sharper. Low confidence = stop, so agents don't act on wrong answers.
โก Same accuracy, 11% shorter reasoning โ faster and cheaper.
๐ https://arxiv.org/abs/2605.14386
๐ค https://huggingface.co/FINAL-Bench/Darwin-180B-RSI
๐๏ธ https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
๐ The result โ #1 on five Hugging Face official leaderboards
๐ฅ AIME 2026 100% (first perfect score on the board)
๐ฅ HMMT Feb 2026 100% (first perfect score on the board)
๐ฅ GPQA Diamond 94.44%
๐ฅ MMLU-Pro 88.12%
๐ฅ MMMU-Pro 79.48%
๐ 131K-token thinking budget ยท bf16 ยท samples per benchmark listed on the model card. ๐
#Darwin #RSI #ZTC #AIME #HMMT #GPQA #MMLUPro #MMMUPro #OpenSourceOrganizations
liked a
Space about 15 hours ago
reacted to SeaWolf-AI's post with โค๏ธ๐ 1 day ago
Post
5142
๐งฌ Darwin-180B-RSI โ an AI that learns from itself and knows when it's right
๐ FINAL-Bench/Darwin-180B-RSI
๐งฌ Darwin โ crossbreed and evolve the parent
Darwin diagnoses strong parent models like an MRI, inherits only their best parts, and evolves the weak spots โ producing a child stronger than its parents.
Father model: Qwen3.8-Flash-Next (180B MoE).
๐ง Rewired paths
๐น 12 full-attention layers ยท ๐น 36 linear-attention layers ยท ๐น 48 shared-expert layers โ precision-strengthened
๐ 512 routed experts ยท router ยท vision encoder โ untouched
โ Only 0.02% of the weights changed.
๐ RSI ร ๐๏ธ ZTC
RSI (recursive self-improvement): solve โ verify against real answers โ learn only the correct reasoning โ repeat.
ZTC (Zero-Token Confidence): reads the model's internal state once, before answering, and returns the probability the answer is right โ zero extra tokens. Returns answer + confidence as JSON.
{"answer": "...", "confidence": 0.97, "truncated": false}
โจ Synergy: ZTC finds where the model wavers โ RSI learns exactly there โ confidence gets sharper. Low confidence = stop, so agents don't act on wrong answers.
โก Same accuracy, 11% shorter reasoning โ faster and cheaper.
๐ https://arxiv.org/abs/2605.14386
๐ค FINAL-Bench/Darwin-180B-RSI
๐๏ธ https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
๐ The result โ #1 on five Hugging Face official leaderboards
๐ฅ AIME 2026 100% (first perfect score on the board)
๐ฅ HMMT Feb 2026 100% (first perfect score on the board)
๐ฅ GPQA Diamond 94.44%
๐ฅ MMLU-Pro 88.12%
๐ฅ MMMU-Pro 79.48%
๐ 131K-token thinking budget ยท bf16 ยท samples per benchmark listed on the model card. ๐
#Darwin #RSI #ZTC #AIME #HMMT #GPQA #MMLUPro #MMMUPro #OpenSource
๐ FINAL-Bench/Darwin-180B-RSI
๐งฌ Darwin โ crossbreed and evolve the parent
Darwin diagnoses strong parent models like an MRI, inherits only their best parts, and evolves the weak spots โ producing a child stronger than its parents.
Father model: Qwen3.8-Flash-Next (180B MoE).
๐ง Rewired paths
๐น 12 full-attention layers ยท ๐น 36 linear-attention layers ยท ๐น 48 shared-expert layers โ precision-strengthened
๐ 512 routed experts ยท router ยท vision encoder โ untouched
โ Only 0.02% of the weights changed.
๐ RSI ร ๐๏ธ ZTC
RSI (recursive self-improvement): solve โ verify against real answers โ learn only the correct reasoning โ repeat.
ZTC (Zero-Token Confidence): reads the model's internal state once, before answering, and returns the probability the answer is right โ zero extra tokens. Returns answer + confidence as JSON.
{"answer": "...", "confidence": 0.97, "truncated": false}
โจ Synergy: ZTC finds where the model wavers โ RSI learns exactly there โ confidence gets sharper. Low confidence = stop, so agents don't act on wrong answers.
โก Same accuracy, 11% shorter reasoning โ faster and cheaper.
๐ https://arxiv.org/abs/2605.14386
๐ค FINAL-Bench/Darwin-180B-RSI
๐๏ธ https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
๐ The result โ #1 on five Hugging Face official leaderboards
๐ฅ AIME 2026 100% (first perfect score on the board)
๐ฅ HMMT Feb 2026 100% (first perfect score on the board)
๐ฅ GPQA Diamond 94.44%
๐ฅ MMLU-Pro 88.12%
๐ฅ MMMU-Pro 79.48%
๐ 131K-token thinking budget ยท bf16 ยท samples per benchmark listed on the model card. ๐
#Darwin #RSI #ZTC #AIME #HMMT #GPQA #MMLUPro #MMMUPro #OpenSource
liked a
model 2 days ago
liked a
model 4 days ago
liked a
Space 5 days ago
liked a
model 5 days ago
upvoted a collection 5 days ago
reacted to SeaWolf-AI's post with ๐๐ฅ 5 days ago
Post
3772
๐ผ๏ธ NO GPU, Only CPU : Z-Image model
Zero graphics cards. 46 seconds. Photoreal.
That laptop you're reading this on. No graphics card, right? It generates images.
No CUDA install. No Python environment. No driver changes. One binary, three model files. Done.
๐ Measured โ GPU count used: zero
512ร512 : 46.4 s
Korean prompt : 45.3 s (faster than English)
1024ร1024 : 192.7 s
Peak RAM : 6.42 GB
GPUs used : 0
(Intel Xeon Gold 6526Y ร2, 48 threads, Q4_0, 3 steps)
โก From 244 seconds to 46 โ 5.3ร
Run it on defaults and it takes 244 s. Switch to 3 steps and it's 48.6 s. Add VAE tiling and it's 46.4 s.
The biggest culprit was the default. Z-Image Turbo is distilled to paint in few strokes, but the tool's default is 20. We were throwing away 5ร for no reason. So were we, at first.
3 is the floor. Put 4 and 3 side by side and you cannot tell them apart. At 2 it collapses โ water droplets and wood grain vanish, and the surface turns cloth-like.
๐ Links
Model
FINAL-Bench/POCKET-Zimage-CPU
Live demo Space (runs on CPU)
FINAL-Bench/POCKET-Zimage-CPU
POCKET collection
FINAL-Bench/pocket-models-6a618ee5d23eafb7e185a5c6
Zero graphics cards. 46 seconds. Photoreal.
That laptop you're reading this on. No graphics card, right? It generates images.
No CUDA install. No Python environment. No driver changes. One binary, three model files. Done.
๐ Measured โ GPU count used: zero
512ร512 : 46.4 s
Korean prompt : 45.3 s (faster than English)
1024ร1024 : 192.7 s
Peak RAM : 6.42 GB
GPUs used : 0
(Intel Xeon Gold 6526Y ร2, 48 threads, Q4_0, 3 steps)
โก From 244 seconds to 46 โ 5.3ร
Run it on defaults and it takes 244 s. Switch to 3 steps and it's 48.6 s. Add VAE tiling and it's 46.4 s.
The biggest culprit was the default. Z-Image Turbo is distilled to paint in few strokes, but the tool's default is 20. We were throwing away 5ร for no reason. So were we, at first.
3 is the floor. Put 4 and 3 side by side and you cannot tell them apart. At 2 it collapses โ water droplets and wood grain vanish, and the surface turns cloth-like.
๐ Links
Model
FINAL-Bench/POCKET-Zimage-CPU
Live demo Space (runs on CPU)
FINAL-Bench/POCKET-Zimage-CPU
POCKET collection
FINAL-Bench/pocket-models-6a618ee5d23eafb7e185a5c6
- 2 replies
replied to SeaWolf-AI's post 6 days ago
One more update to the board. JEV's latency is now the same-place round trip, 0.591 s (ZTC 0.636 s), replacing the 2.1 s figure. With that, ZTC and JEV tie in the same 60 s too (-0.07 lines [-0.14, +0.01]); ZTC over the always-re-roll arm stays +1.03 [0.79, 1.26]. Laya is wired at 0.4904, measured on the corrected set.
So the board separates on judging, and against JEV the difference is where it runs: zero generated tokens, local weights, no network hop.
replied to SeaWolf-AI's post 6 days ago
Good catch on the first draw. It now sits at 0.559, the rate the re-draw chain settles at, so a re-roll without judging buys nothing. At its measured 0.50 the Laya board is the always-re-roll null arm, and the page labels it that way.
Re-run on the updated board, 1,000 paired matches, same 60 s: ZTC minus always-re-roll +1.06 lines [0.82, 1.30], ZTC minus JEV +4.16 [4.01, 4.33]. Same moves: ZTC and JEV tie.
On your question: a re-sample shouldn't beat the first sample on its own. The measured 38% / 30% transition rates imply the first draw also sits at 0.559, which is what the board uses now.
liked a
Space 7 days ago
reacted to SeaWolf-AI's post with ๐ฅ 7 days ago
Post
3398
The cost of a judging gate is usually quoted as a number. This puts it on a Tetris board.
Three boards get the same piece order, and on every move the same proposal and the same noise โ a paired comparison. The gate decides one thing: keep this move, or draw again. Each board gets the same 60 seconds of gate time.
The text-writing gates get through 15โ22 moves. The generation-free gate gets through 40โ50. The boards that stop simply run out of clock.
It does not win on accuracy: on the same 2,018-question LODO set, JEV scores AUC 0.7350 against ZTC-Judge-27B's 0.7289. The separation is elsewhere. Clock โ 2.1 s vs 0.0615 s per call, and on a 200-candidate agent screen one judging call measured 3.206 s generative vs 0.033 s readout, same server. Calibration โ a gate is a threshold, and at ECE 0.4985 (vs ZTC 0.0245) a threshold stops carrying information. Mechanism โ a text judge can name option 42 when there is no option 42; a scoring readout cannot. Not a lower error rate. No path.
The curve in the ZTC panel is real online fitting, scored prequentially โ predict first, learn after โ with base weights untouched. Not recursive self-improvement.
Limits, also stated on the page: Laya's AUC and latency are not our measurements and are set equal to JEV's, so calibration is the only measured axis it differs on. The page is a simulation driven by measured constants.
KO / EN / ZH.
FINAL-Bench/Tetris-JEV-LAYA-ZTC
FINAL-Bench/ZTC-Judge-27B
Three boards get the same piece order, and on every move the same proposal and the same noise โ a paired comparison. The gate decides one thing: keep this move, or draw again. Each board gets the same 60 seconds of gate time.
The text-writing gates get through 15โ22 moves. The generation-free gate gets through 40โ50. The boards that stop simply run out of clock.
It does not win on accuracy: on the same 2,018-question LODO set, JEV scores AUC 0.7350 against ZTC-Judge-27B's 0.7289. The separation is elsewhere. Clock โ 2.1 s vs 0.0615 s per call, and on a 200-candidate agent screen one judging call measured 3.206 s generative vs 0.033 s readout, same server. Calibration โ a gate is a threshold, and at ECE 0.4985 (vs ZTC 0.0245) a threshold stops carrying information. Mechanism โ a text judge can name option 42 when there is no option 42; a scoring readout cannot. Not a lower error rate. No path.
The curve in the ZTC panel is real online fitting, scored prequentially โ predict first, learn after โ with base weights untouched. Not recursive self-improvement.
Limits, also stated on the page: Laya's AUC and latency are not our measurements and are set equal to JEV's, so calibration is the only measured axis it differs on. The page is a simulation driven by measured constants.
KO / EN / ZH.
FINAL-Bench/Tetris-JEV-LAYA-ZTC
FINAL-Bench/ZTC-Judge-27B
- 5 replies
liked 2
models 9 days ago
reacted to SeaWolf-AI's post with ๐๐ฅ 9 days ago
Post
3164
Ask a language model how confident it is and you get an AUC of 0.5000. Exactly a coin flip. We measured it across 2,018 items.
FINAL-Bench/gate-tetris
https://huggingface.co/blog/FINAL-Bench/ztc
Collection: https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
Zero-Token Confidence (ZTC) reads it. One forward pass over the model's hidden state returns a calibrated probability that the answer is correct. Zero generated tokens.
It sits at the top of the shared board. Same 2,018 items, same harness for every entry: ZTC on Darwin-397B 0.7394, JEV 0.7335, ZTC-Judge-27B 0.7255, a surface baseline that reads only answer length and formatting 0.7036, Lynx 8B 0.5157, the model's own self-reported confidence 0.5000, HHEM 0.4852. First and third place both emit nothing at all.
The number worth staring at is 0.7036. That is a baseline reading no content whatsoever, just how long the answer is and how it is formatted. Any verifier scoring below it is not reading content either.
On speed, one gate call costs 0.0615 seconds, measured on four B200s across 2,000 items. Generating a single candidate answer takes 1.631 seconds, so the gate is 26 times cheaper than the work it guards. A verifier that generates competes with your agent for the same budget. A verifier that only reads can be attached to every action instead of a sampled few.
We built it so you can watch it decide. Three lanes receive the same stream of proposed actions and the same time budget. One has no gate and must execute everything. One uses a text-reading verifier. One uses ZTC. Right action plus one, wrong action minus one, hold zero. Over 400 matches: no gate minus 3.9, text verifier plus 13.0, ZTC plus 29.1, with ZTC taking 98 percent of matches. Gating lifts executed accuracy from 49 percent to 65 percent.
FINAL-Bench/gate-tetris
https://huggingface.co/blog/FINAL-Bench/ztc
Collection: https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems
Zero-Token Confidence (ZTC) reads it. One forward pass over the model's hidden state returns a calibrated probability that the answer is correct. Zero generated tokens.
It sits at the top of the shared board. Same 2,018 items, same harness for every entry: ZTC on Darwin-397B 0.7394, JEV 0.7335, ZTC-Judge-27B 0.7255, a surface baseline that reads only answer length and formatting 0.7036, Lynx 8B 0.5157, the model's own self-reported confidence 0.5000, HHEM 0.4852. First and third place both emit nothing at all.
The number worth staring at is 0.7036. That is a baseline reading no content whatsoever, just how long the answer is and how it is formatted. Any verifier scoring below it is not reading content either.
On speed, one gate call costs 0.0615 seconds, measured on four B200s across 2,000 items. Generating a single candidate answer takes 1.631 seconds, so the gate is 26 times cheaper than the work it guards. A verifier that generates competes with your agent for the same budget. A verifier that only reads can be attached to every action instead of a sampled few.
We built it so you can watch it decide. Three lanes receive the same stream of proposed actions and the same time budget. One has no gate and must execute everything. One uses a text-reading verifier. One uses ZTC. Right action plus one, wrong action minus one, hold zero. Over 400 matches: no gate minus 3.9, text verifier plus 13.0, ZTC plus 29.1, with ZTC taking 98 percent of matches. Gating lifts executed accuracy from 49 percent to 65 percent.
liked a
Space 9 days ago
liked a
model 10 days ago