chi-bench: Actava · openai-agents · nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16:peft:262144 - #21
Merged
haolin-chen-actava merged 1 commit intoAug 12, 2026
Conversation
…0B-A55B-BF16:peft:262144 Filed 2026-08-12 from the original 2026-07-22/24 run; aggregation unmodified. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Submission validation —
|
haolin-chen-actava
deleted the
sub/chi-bench/2026-07-24-nemotron-3-ultra-openai-agents
branch
August 12, 2026 20:20
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
NVIDIA Nemotron 3 Ultra 256K on the OpenAI Agents SDK harness via Tinker Chat Completions, with an explicit
provider_route: tinker.Full-run cost $35.42 (limited-time Tinker discount verified 2026-07-22).
This is a genuine capability result, not a harness failure. Mean fractional reward is 0.217 — the model produced partial task progress throughout but never satisfied a full rubric. The evaluation was gated before and during the run:
result.jsonandverifier/scorecard.jsonMaxTurnsExceededoutcome, which is a genuine agent result rather than a harness faultClosed per
docs/superpowers/plans/2026-07-22-nemotron-tinker-eval.md.Trajectory repair. The one
MaxTurnsExceededtrial completed verification but emitted notrace.jsonl; an error-only ATIF-v1.2 trajectory was generated with the officialopenai_agents_harness._build_atif_trajectory(records=[], ...)builder. Original logs unmodified.chi_bench_git_shapins2395e9e, now onmainvia actava-ai/chi-bench#42 — which also carries the Nemotron Tinker tool-protocol adapter this run depends on.Validator: 0 errors, 0 warnings.
🤖 Generated with Claude Code