{"id":"6db0eeb8-1f65-4bc3-8ed4-c42f48adf062","arxiv_id":"2610.28480","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A single learned differentiable world model lets one GPU train a 129,514-motion humanoid whole-body tracker in 3 hours, surpassing a 21,000 GPU-hour PPO baseline and transferring to a real Unitree G1.","lead":"This paper trains a humanoid whole-body motion controller by learning a differentiable model of the robot's shared dynamics, then improving the controller with gradients through short imagined rollouts. On one GPU it matches or beats a cluster-trained baseline in hours instead of weeks, and the resulting policy works on a physical Unitree G1.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Failure counts are scored only under SuperMimic's own termination rule; Appendix F drops SONIC's end-effector criterion, so the 25-vs-179 gap and zero-failure claim are not yet shown to be metric-independent. A two-criterion re-scoring would settle it.","rationale":"Good-faith read: the method is coherent, the LAFAN1 PPO-Sim comparison is well controlled, the compute accounting is transparent, and the limitations section states the main scope restrictions. The strongest independent support is the PPO-Sim experiment, which holds observation, reward, termination, domain randomization, network sizes, and evaluation protocol fixed and still shows SuperMimic reaching 0.82 in 1.2 h while PPO-Sim stays below 0.72 after 18 h on up to eight GPUs. That supports the efficiency claim without relying on the SONIC comparison. The reader's weakest assumption points at SONIC comparison fairness; I partially agree. The filtered benchmark is less suspect because the name filter is adapted from SONIC's own released data-processing code and the feasibility thresholds are disclosed, and the GPU-type gap is bounded by the authors' stated 5.5x peak-throughput note. The narrowest load-bearing issue is Appendix F's failure metric: the counts are generated under SuperMimic's termination only, and the paper explicitly states that SONIC's own criterion differs. Because the abstract's failure numbers depend on this choice and no alternative counts are given, the failure part of the central claim is not yet established as metric-independent. Re-scoring under both criteria would settle it without altering the method. The verdict stays CONDITIONAL; if both criteria give the same ordering, the result should move toward ACCEPT.","tokens_in":19200,"tokens_out":14813,"duration_ms":115656,"concrete_test":"Re-run the Appendix F scoring for the released SONIC checkpoint, SONIC v1.1, and the SuperMimic 3 h/one-GPU and 9.8 h/two-GPU checkpoints on the same 129,514 motions, and report failure counts under three definitions: (i) SuperMimic's height/tilt termination; (ii) SONIC's published success criterion (root and end-effector height deviation above 0.25 m, no orientation term); and (iii) a neutral rule that triggers on either criterion. Also report median normalized return under each rule. If SONIC is not consistently worse under (ii) and (iii), the failure-gap and zero-failure claims are evaluation-specific; if the ordering survives both criteria, the concern is resolved and the headline stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing soft spot is Appendix F's failure definition, not the method itself. The abstract and Section 4.1 advertise 25 versus 179 failures on 129,514 motions and zero failures after 9.8 h, but every failure count in Appendix F is computed under a single rule: a motion fails if its rollout triggers SuperMimic's training termination. That termination is SuperMimic's height/tilt threshold, and Appendix A explicitly drops the end-effector termination. SONIC's published success criterion differs, failing motions on root and end-effector height deviations above 0.25 m without an orientation term. The counts are also sensitive to protocol choice: the released SONIC checkpoint gives 179 failures and the v1.1 checkpoint gives 239 under the same SuperMimic rule. No counts are reported under SONIC's own criterion or under a neutral combined criterion, so the abstract's failure gap and zero-failure claim are not yet shown to be properties of the policies rather than of the chosen failure definition. The return-based >1000x speedup is on firmer ground because it is independently supported by the controlled LAFAN1 PPO-Sim comparison and by the authors' own compute accounting, including the stated 2.3x data difference and at most 5.5x peak-GPU-throughput gap. The failure-count claim is the least secure pillar of the headline.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"This paper proposes SuperMimic, a model-based RL pipeline for humanoid whole-body motion tracking. It trains a single differentiable neural world model on logged MuJoCo Warp transitions, conditions it on per-episode simulation context, and uses backpropagation-through-time over H=32 imagined rollouts, with a critic bootstrapping the return beyond the horizon, to train a goal-conditioned policy. On a filtered 129,514-motion BONES-SEED benchmark, the paper reports that SuperMimic surpasses the return of the released SONIC checkpoint within 3 hours on one H100 GPU, reduces GPU-hours by more than 1000x, fails on 25 motions versus 179 for SONIC, reaches zero failures after 9.8 hours on two GPUs, keeps world-model prediction error stable as data grows from 1 to 264 hours, transfers to a physical Unitree G1, and outperforms a distillation baseline with an RL-trained velocity encoder. A controlled LAFAN1 comparison against PPO in the simulator, exact hyperparameters, and detailed compute accounting are provided in support of the efficiency claims.","tokens_in":19689,"tokens_out":6772,"duration_ms":48481,"significance":"If the headline result holds, this is a substantial advance: it would make cluster-scale compute unnecessary for whole-body motion tracking and would make scaling-law studies of motion tracking affordable on a single GPU. The paper is unusually careful in several places: exact hyperparameters are given in Table A.5, the training loop is fully specified in Algorithm A.1, and the compute accounting explicitly quantifies the 2.3x data difference and the at-most-5.5x peak-GPU-throughput difference between the H100 and L40. The controlled LAFAN1 PPO-Sim comparison is also a genuine strength, since it controls motions, observations, reward, termination, domain randomization, network sizes, and evaluation protocol. The usual learned-simulator circularity concern does not land here in its standard form: the world model is fit to logged simulator transitions, and the main efficiency comparison in Figure 4 is against a model-free PPO policy evaluated in the same simulator. The main unresolved issue is whether the advertised failure-count advantage reflects the policies rather than the chosen failure definition; the return-based speedup itself is on substantially firmer ground.","major_comments":[{"comment":"The failure-count headline, 25 versus 179 failures on 129,514 motions and zero failures after 9.8 hours, is computed in Appendix F only under SuperMimic's termination rule. Appendix A defines that rule as root height/tilt termination (|e_p,z| > 0.25, tilt error > 0.8) and explicitly drops BeyondMimic's end-effector termination, while Appendix F states that SONIC's own success criterion fails motions on root and end-effector height deviations above 0.25 m without an orientation term. The v1.1 SONIC checkpoint also moves from 179 to 239 failures under the same SuperMimic rule, so the claimed gap is not yet shown to be a property of the policies rather than of the chosen failure definition. I request re-scoring of the SONIC released checkpoint, the SONIC v1.1 checkpoint, and the SuperMimic checkpoints under SONIC's own criterion and under a combined \"either criterion\" rule, with both counts reported in Table A.8 or in the main text.","section":"§4.1, Appendix F, Appendix A","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The failure-count comparison is also sensitive to the benchmark construction and checkpoint choice. Appendix E removes 12,687 clips using name and feasibility thresholds that are stated to be tuned for the G1 retarget at 50 Hz, and Appendix F evaluates only the filtered set with no domain randomization or observation noise. Because the same author-chosen pipeline filters the motions, sets the scoring termination, and selects the released checkpoint for the headline, the 25-versus-179 gap should be accompanied by a robustness table varying the feasibility thresholds and reporting both SONIC checkpoints. Without such a table, the main text overstates the certainty of the failure-count advantage, although the return-based compute speedup is not affected by this concern.","section":"Appendix E, Appendix F","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The abstract and Section 4.1 describe the baseline as \"a state-of-the-art motion tracker\" but the reported failures are for the released checkpoint only; stating the checkpoint identity in the main text would avoid the implication that all SONIC variants behave identically.","section":"Abstract and §4.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Appendix F correctly notes that the benchmark motions are training motions for both methods and therefore measure training coverage, not generalization; this caveat should appear in the main text next to the failure counts, since readers may otherwise interpret them as generalization results.","section":"Appendix F","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The sentence claiming that \"from then on, every checkpoint is ahead of SONIC by at least half of the gap\" is not backed by a visible curve or table; either show the margins for all checkpoints after 3 hours or state a precise checkpoint-by-checkpoint rule.","section":"Figure 3(a) and §4.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The statement that the learned model \"remains stable over indefinitely long rollouts\" is supported only by the k-step autoregressive fitting argument; a curtailed rollout-stability curve or a quantitative statement of the longest verified rollout would make this claim easier to check.","section":"§3.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Appendix G states that each velocity-encoder policy has one training seed; this should be stated in Section 4.5 as well, so that the 650/672 result is not read as a multi-seed comparison.","section":"§4.5 and Appendix G","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The headline is mostly true, but the failure counts are the part to be careful with. What is actually new: one context-conditioned differentiable world model amortizes a 129,514-motion humanoid library, and policy learning via BPTT through imagined rollouts reaches SONIC-level return on a single GPU in hours rather than 21,000 GPU-hours. That is a real result, and the direction of the efficiency claim is solid. The LAFAN1 controlled comparison is the strongest evidence: PPO in the simulator sharing motions, observations, reward, termination, domain randomization, and network sizes never reaches return 0.82 in 18 hours on 8 GPUs, while SuperMimic does in 1.2 hours on one. The compute accounting is unusually honest, stating the 2.3x data gap and the H100/L40 peak-throughput ratio themselves. The ablations earn their place: 64x fewer imagined steps than PPO-on-world-model, and the critic ablation showing shorter horizons and push recovery.\n\nThe soft spot is exactly where the stress test points. Every failure count in Appendix F is scored under SuperMimic's own termination, which explicitly drops the end-effector criterion (Appendix A says so). SONIC's published criterion differs, so the 25-vs-179 gap and the zero-failure claim are not yet shown to be properties of the policies rather than of the chosen rule. The abstract states them without that caveat, and that is the one place the paper overreaches. The fix is cheap: re-score all checkpoints under SONIC's criterion and under a combined rule. I would also note the benchmark motions are training motions for both methods, so these counts measure training coverage, not generalization; the paper says this itself in F.2, so it is an acknowledged scope limit, not a hidden flaw. Secondary issues: no released code or per-seed error bars (\"training very stable\" is asserted, not shown), and the real-robot section is four qualitative clips.\n\nI do not think the comparison is rigged, and this is not curve-fitting to the target: the world model learns simulator dynamics, and the returns, failures, and speedups are measured outputs. The Appendix E filter removes 12.7k clips with thresholds tuned for the G1 retarget, but both methods train and are scored on the filtered set, so it is symmetric enough to survive scrutiny.\n\nThis deserves a serious referee. The result is important if it holds, the method section is detailed enough to check, and the comparison-protocol question is exactly what referees should adjudicate. Send it out, and ask for the two-criterion re-scoring plus artifacts.","headline":"The efficiency claim is real and well controlled; the 25-vs-179 failure numbers rest on SuperMimic's own termination rule and should be re-scored under SONIC's criterion before quoting them.","tokens_in":20900,"tokens_out":722,"duration_ms":26802},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"SuperMimic claims that one differentiable world model of the robot's shared dynamics, fit from simulator log data and improved together with the policy, lets a single GPU beat a 128-GPU, 7-day-trained motion tracker's return in 3 hours…","keywords":["humanoid whole-body control","differentiable world model","model-based reinforcement learning","motion imitation","backpropagation-through-time","analytic policy gradients","sim-to-real transfer","single-GPU training"],"falsifier":"Run the benchmark both ways: score SONIC's released checkpoint under its own end-effector-inclusive termination and under SuperMimic's rule, and train a PPO pipeline from scratch on SuperMimic's exact filtered data, reward, termination, and network sizes within a comparable GPU-hours budget; if PPO reaches SONIC's return in comparable compute, or if SONIC's failure count under its own criterion drops below SuperMimic's 25, the headline gaps overstate the advantage. Separately, fit the world model on a motion library well beyond 264 hours, or on the terrain and object-contact motions the paper excludes; if 8-step prediction error grows with library size or motion diversity, the claim that one model amortizes all motions is scoped to flat-ground data.","tokens_in":20707,"feed_emoji":"🤖","tokens_out":11794,"duration_ms":74242,"temperature":0.7,"pith_summary":"SuperMimic sets out to show that training compute is no longer the bottleneck to scaling humanoid whole-body control. Its proposal is to fit one differentiable world model to logged transitions of the robot's rigid-body simulator, a single network that captures the dynamics every reference motion shares, and to train the tracking policy by backpropagating analytic gradients through short imagined rollouts of that model instead of estimating gradients from noisy simulator rollouts as model-free reinforcement learning does. If the central claim holds, the same model amortizes learning across an entire motion library: prediction error stays flat as the training data grows from 1 to 264 hours, one H100 surpasses the return of a policy trained on 128 GPUs for 7 days within 3 hours (a more than 1000-fold reduction in GPU-hours), and failures drop to 25 versus 179 on a shared 129,514-motion benchmark, with none remaining after 9.8 hours on two GPUs. The resulting controller transfers to a physical Unitree G1 across a skipping jump, a side roll, and two flips, and the same dynamics model trains a downstream velocity interface on its own task reward.","feed_headline":"One GPU beats a 128-GPU humanoid trainer in 3 hours","feed_subtitle":"A single learned dynamics model cuts GPU-hours 1000-fold and fails 25 of 129,514 motions versus the baseline's 179.","key_machinery":"The load-bearing object is the differentiable world model $f_\\omega$: a deterministic MLP, conditioned on state, action, and the per-episode simulator context $\\xi$, with learned reward, observation, and termination heads whose outputs keep the survival weight differentiable in imagined rollouts. It is fit on $k=8$-step windows drawn from a replay buffer the policy itself fills, and the policy update backpropagates an $H=32$-step imagined return through it, seeded from replay states in the Dyna style, with a critic $V_\\psi$ trained on GAE targets that bootstraps the return beyond the horizon, a short-horizon actor-critic in the spirit of SHAC. The argument rests on three measured properties of this model: its 8-step prediction error stays within 0.007-0.011 as the motion library grows from 1 to 264 hours, its rollouts run more than 10 times faster than the simulator at the batch sizes used (up to 62 times at 64k environments), and its analytic gradients reach the best return achieved by PPO trained on world-model rollouts with about 64 times fewer imagined steps.","core_discovery":"The paper's central claim is that a whole-body motion-tracking policy for a humanoid can be trained by learning a single differentiable surrogate of the simulator and differentiating an imagined return through it, rather than by the model-free policy-gradient estimates that have dominated this setting. The world model $f_\\omega$ predicts next state, reward, observation, and termination from the current state, the executed action, and a per-episode simulator context $\\xi$ covering friction, calibration offsets, and torso center-of-mass; it is fit autoregressively on short replay windows collected by the policy itself, so dynamics learning and control improve in one loop. Because only the learned model needs gradients, the approach stays compatible with rigid-body simulators whose contacts are non-smooth and whose implementations are non-differentiable, and the first-order gradients carry much lower variance than likelihood-ratio estimates over short horizons. Concretely, the released SONIC checkpoint, trained for roughly 21,000 GPU-hours, is surpassed in normalized return within 3 hours on one H100; on the shared 129,514-motion benchmark SuperMimic fails on 25 motions after 3 hours on one GPU and on none after 9.8 hours on two, versus 179 for the baseline; and on a controlled LAFAN1 comparison, the return that PPO-in-simulator never reaches in 18 hours on up to eight GPUs is reached in 1.2 hours on one.","pith_inferences":["If the flat prediction-error trend continues beyond 264 hours, the amortization argument should extend to other embodiments and to wholly new skill classes without retraining the dynamics; measuring 8-step prediction error on a much larger library, or on a second robot, would test that transfer.","The failure gap is sensitive to evaluation rules: both policies are scored under SuperMimic's height-and-tilt termination, which omits the baseline's end-effector criterion, so re-running the 129,514-motion benchmark under both termination rules would show how much of the 179-to-25 gap is rule choice.","Because the world model is differentiable and conditions on simulator context, it is a natural substrate for online system identification or fine-tuning on real-robot data; the paper notes the model currently learns only from simulator transitions, so exploiting that differentiability is a direct next step.","The flat error curve implies a testable scaling law: on a fixed humanoid, GPU-hours to a target tracking return should grow only with the cost of covering rarer motions, not with model capacity, and running the sweep to 1,000 or more hours of motion would show whether the trend survives."],"forward_implications":["Training compute stops being the limiting resource for whole-body motion tracking: scaling laws for motion-tracking controllers can now be measured at a few GPU-hours per run instead of tens of thousands.","Model-free PPO is not required for deployable humanoid control: backpropagation through a learned surrogate trains policies that transfer to hardware even when the simulator itself is not differentiable.","Adding motions to the library should stay cheap: since prediction error does not grow from 1 to 264 hours of data, one world model absorbs new clips without demanding new capacity or new compute.","One dynamics model can serve several goal interfaces at once: the same world model trains the motion tracker and a downstream velocity controller, including directly on the interface's own task reward rather than by distillation.","The critic is what makes short imagination horizons viable: without it, training fails at horizons of 16 steps or less, and with it, the policy recovers from velocity pushes in 0.5-2.4 seconds where the critic-free policy never settles."],"supporting_citations":[{"why":"Supplies SONIC, the released cluster-scale PPO checkpoint and its roughly 21,000 GPU-hour training cost that define the comparison baseline for return, failures, and compute.","marker":"[1]"},{"why":"Supplies the BeyondMimic motion-tracking reward, termination, goal conditioning, and PPO hyperparameters that both SuperMimic and the PPO-Sim control baseline build on.","marker":"[3]"},{"why":"Supplies the SHAC short-horizon actor-critic template, backpropagation through differentiable dynamics with a bootstrapping critic, which SuperMimic adapts to a learned surrogate with a learned termination.","marker":"[19]"},{"why":"Supplies MuJoCo, the rigid-body simulator whose logged transitions fit the world model and whose GPU implementation is the rollout-speed baseline.","marker":"[26]"},{"why":"Supplies BONES-SEED, the public motion dataset from which the 264.56-hour filtered training set and the 129,514-motion benchmark are drawn.","marker":"[27]"},{"why":"Supplies LAFAN1, the dataset for the controlled one-GPU-versus-multi-GPU PPO-Sim comparison that isolates training efficiency from benchmark selection.","marker":"[28]"},{"why":"Supplies the generalized advantage estimation recursion used to compute the critic's fixed targets within the imagination window.","marker":"[25]"},{"why":"Supplies SuperTrack, the learned differentiable surrogate for motion-control policy optimization that SuperMimic's analytic-gradient approach scales to large humanoid whole-body tracking.","marker":"[12]"}],"fun_headline_variants":["One GPU, 3 hours: humanoid whole-body control beats 128-GPU baseline","Single differentiable world model cuts humanoid training compute 1000x","After 9.8 hours on two GPUs, SuperMimic fails zero of 129,514 motions","SuperMimic scales humanoid motion control with one learned simulator"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The reported more-than-1000x speedup assumes that SONIC's released checkpoint, scored on a benchmark whose filtering thresholds are tuned for the G1 retarget and whose failure rule is SuperMimic's height-and-tilt termination, is a fair and representative measure of what cluster-scale PPO with the same reward achieves, though the controlled LAFAN1 comparison against PPO trained from scratch gives the efficiency claim independent support.","fun_headline_variants_meta":{"raw":{"variants":["One GPU, 3 hours: humanoid whole-body control beats 128-GPU baseline","Single differentiable world model cuts humanoid training compute 1000x","After 9.8 hours on two GPUs, SuperMimic fails zero of 129,514 motions","SuperMimic scales humanoid motion control with one learned simulator"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.00031,"raw_usage":{"cost":0.0003100006,"is_byok":false,"cost_details":{"upstream_inference_cost":0.0003100006,"upstream_inference_prompt_cost":3.33606e-05,"upstream_inference_completions_cost":0.00027664},"total_tokens":2725,"prompt_tokens":1205,"completion_tokens":1520,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":1048,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":1441}},"tokens_in":1205,"tokens_out":1520,"cost_source":"openrouter_usage","duration_ms":11989,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":1441,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-21T04:02:20.583867+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Run the benchmark both ways: score SONIC's released checkpoint under its own end-effector-inclusive termination and under SuperMimic's rule, and train a PPO pipeline from scratch on SuperMimic's exact filtered data, reward, termination, and network sizes within a comparable GPU-hours budget; if PPO reaches SONIC's return in comparable compute, or if SONIC's failure count under its own criterion drops below SuperMimic's 25, the headline gaps overstate the advantage. Separately, fit the world model on a motion library well beyond 264 hours, or on the terrain and object-contact motions the paper excludes; if 8-step prediction error grows with library size or motion diversity, the claim that one model amortizes all motions is scoped to flat-ground data.","supporting_citations":[{"cited_title":"Robust motion in-betweening","cited_arxiv_id":null,"evidence_quote":"Supplies LAFAN1, the dataset for the controlled one-GPU-versus-multi-GPU PPO-Sim comparison that isolates training efficiency from benchmark selection."},{"cited_title":"High-Dimensional Continuous Control Using Generalized Advantage Estimation","cited_arxiv_id":"1506.02438","evidence_quote":"Supplies the generalized advantage estimation recursion used to compute the critic's fixed targets within the imagination window."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies SONIC, the released cluster-scale PPO checkpoint and its roughly 21,000 GPU-hour training cost that define the comparison baseline for return, failures, and compute."},{"cited_title":"BeyondMimic: From motion tracking to versatile humanoid control via guided diffusion,","cited_arxiv_id":null,"evidence_quote":"Supplies the BeyondMimic motion-tracking reward, termination, goal conditioning, and PPO hyperparameters that both SuperMimic and the PPO-Sim control baseline build on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the SHAC short-horizon actor-critic template, backpropagation through differentiable dynamics with a bootstrapping critic, which SuperMimic adapts to a learned surrogate with a learned termination."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies MuJoCo, the rigid-body simulator whose logged transitions fit the world model and whose GPU implementation is the rollout-speed baseline."},{"cited_title":"BONES-SEED: Skeletal everyday embodiment dataset, 2026","cited_arxiv_id":null,"evidence_quote":"Supplies BONES-SEED, the public motion dataset from which the 264.56-hour filtered training set and the 129,514-motion benchmark are drawn."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies SuperTrack, the learned differentiable surrogate for motion-control policy optimization that SuperMimic's analytic-gradient approach scales to large humanoid whole-body tracking."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}