pith:IOAUKH62
Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy.
arxiv:2601.03969 v2 · 2026-01-07 · cs.AI · cs.CL
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{IOAUKH622IANBKDQEPR6F55X5X}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.
That selectively truncating only the extreme tail of response lengths within correct rollout groups will not degrade the model's ability to learn long-horizon reasoning on complex problems.
Dynamic Outlier Truncation during training reduces token usage by 78% on AIME-24 while increasing accuracy by suppressing extreme-length correct rollouts and adding KL regularization.
Formal links
Receipt and verification
| First computed | 2026-05-17T23:39:16.699768Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "d957aed52f06d32163ff066e42dbe1c0311372ef4c98a1318662217d0246d440",
"cross_cats_sorted": [
"cs.CL"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.AI",
"submitted_at": "2026-01-07T14:31:07Z",
"title_canon_sha256": "dbdd4b9df4587477ef3565737c066a1bbc88c4d7321146d29860b177aea0af81"
},
"schema_version": "1.0",
"source": {
"id": "2601.03969",
"kind": "arxiv",
"version": 2
}
}