Pith. sign in

REVIEW 4 major objections 6 minor 36 references

SPyCE: Skill-Policy Co-evolution for Multimodal Agents

T0 review · 4 major / 6 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SPyCE claims that distilling trajectories into an evolving hierarchical skill library makes multimodal agents outperform both reward-only reinforcement learning and static-memory retrieval.

desk verdict A coherent and plausibly novel training recipe for multimodal tool-use agents, but the reported evidence is point estimates only and the reward is not aligned across baselines—worth a serious referee, not yet a proven claim. read the letter →

arxiv 2607.13854 v1 pith:U3ZZYYWM submitted 2026-07-15 cs.CL

Ru Zhang , Weijie Qiu This is my paper
classification cs.CL
keywords multimodalagentsskill-policyco-evolutionhierarchicalskilllibraryreinforcementlearningtoolusevisualreasoningtrajectorydistillationGRPO
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

SPyCE tries to show that a multimodal agent's own successful trajectories should be turned into reusable skills during training, rather than thrown away as scalar rewards or stored as static memories. The paper builds a two-level skill library — short execution skills and longer workflow skills — and continuously updates it from the policy's best rollouts while the policy trains on retrieved skills. The claim is that this closed loop beats both RL-only and memory-only alternatives, with gains of roughly one to four accuracy points across eight benchmarks, fewer tool calls, and faster early convergence. A reader should care because it reframes experience itself as a training signal: the agent gets better, and the better agent's traces make the next round of skills better.

What carries the argument

The load-bearing object is the two-level skill library plus its update loop. Execution skills are triples (c, u, e) — a trigger condition, a tool action, and an expected effect — retrieved by embedding similarity on the condition. Workflow skills are pairs (b, w) — a coarse bottleneck summary and a workflow sketch — retrieved by similarity on the bottleneck. Retrieval is hierarchical: the policy first writes a bottleneck query, pulls workflow skills, then decomposes the task into local bottleneck descriptors and pulls one execution skill per descriptor. A merge-or-add consolidation rule merges near-duplicate candidates and prunes the least-used entries, keeping both libraries bounded. The po

What would settle it

Run the same training recipe with a skill library built once by the same teacher model and then frozen, keeping the identical reward, retrieval procedure, and rollout budget. If the frozen-library version matches or exceeds the online co-evolution version on the agentic benchmark, the claim that co-evolution is critical is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hierarchical skill library updated in lockstep with the policy produces measurably better multimodal tool-use agents. Concretely, on the reported benchmarks the co-evolved agent beats plain group-relative policy optimization by 0.8 to 4.3 accuracy points, and on an agentic reasoning benchmark it reaches 32.0 average accuracy with 4.83 average tool calls, compared with 27.7 and 5.13 for GRPO. The mechanism is a closed loop: execution skills (trigger condition, tool action, expected visual effect) are matched to local bottlenecks; workflow skills (bottleneck summary, workflow sketch) guide multi-step planning; the policy conditions on both while optimizing t

Load-bearing premise

The load-bearing premise is that the online co-evolution update, rather than the teacher's distilled skill priors or the changed reward, is what produces the final-points gain; the paper does not run a frozen-teacher-library control under the identical reward.

Editorial extensions

If this is right

  • If the central claim holds, trajectory distillation into skills should be adopted alongside policy-gradient training for multimodal tool-use agents, not treated as an alternative to it.
  • The reported tool-call counts imply that skill conditioning makes agents more selective about tools, not merely more eager to use them.
  • The ablation outcomes imply that neither skill level alone is sufficient: workflow-level guidance and execution-level precision contribute separately, and freezing the library costs about three points.
  • The tool-call penalty results imply that a penalty applied only to successful episodes can improve both accuracy and efficiency, whereas a blanket penalty suppresses useful exploration.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A clean next experiment would compare the online co-evolved library against a static library distilled by the same teacher under the identical reward; that isolates the update loop from the teacher's priors.
  • Because skills are stored as text, one could transfer a library trained with one backbone to a smaller or larger backbone at inference, potentially turning training-time co-evolution into test-time skill reuse.
  • The observed early convergence suggests the skill library acts partly as a curriculum, shaping exploration toward bottlenecks that recur across tasks; testing whether skill diversity correlates with exploration would connect this to exploration-exploitation theory.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes SPyCE, a framework for multimodal tool-use agents that couples GRPO-style reinforcement learning with a hierarchical skill library. During training, successful rollout trajectories are distilled by a large teacher MLLM (Qwen3-VL-235B) into execution skills (condition–action–effect triples) and workflow skills (bottleneck description plus workflow sketch). The policy conditions on retrieved skills during rollouts, and the libraries are periodically updated from high-quality rollouts, creating a claimed closed skill-policy co-evolution loop. Experiments on eight benchmarks (TIR-Bench, V*, HRBench-4K/8K, ChartQAPro, MathVerse, MathVision, WeMath) report that SPyCE outperforms prompt-based, RL-based (GRPO, RLOO), and memory-based (MemP, Dynamic Cheatsheet, Agent-KB) baselines. Ablations on TIR-Bench attribute gains to the hierarchical skill design, online skill evolution, and the tool-call penalty.

Significance. If the reported results are robust, the paper makes a worthwhile contribution: it addresses an important limitation of current RL-trained multimodal agents—trajectories are reduced to scalar rewards rather than reused as structured knowledge—and the proposed closed-loop skill-policy co-evolution is a timely idea. The evaluation covers several public benchmarks and compares with both RL and memory-augmented baselines, and the ablations attempt to isolate the contributions of the two skill levels and online evolution. The convergence and tool-use analysis is also informative. However, the central claims currently rest on single-run numbers with no variance estimates, and several comparisons have margins small enough to be within plausible run-to-run noise. The reward used for the RL baselines is not specified, which is a direct confound for the headline comparisons. These issues are fixable but currently block acceptance.

major comments (4)
  1. [§4.2, Tables 1–2] No standard deviations, seeds, or significance tests are reported. Several claimed gains over the strongest baseline are small: HRBench-4K +0.8, ChartQAPro +0.9, WeMath +1.2, HRBench-8K +1.4, V* +2.1 (8B backbone). Without a distribution over runs, the statement 'consistently outperforms' is not statistically supported. Please report the number of runs, mean ± std, and a significance test where appropriate.
  2. [§4.1.2 and Eq. (3)] The reward function used for the GRPO and RLOO baselines is not specified. SPyCE uses R(τ)=R_acc(τ)−λ_call·1[Succ]·N_tool, with λ_call=0.05. If the baselines were trained with a different reward (e.g., accuracy only, or a different penalty), the comparison is confounded: the gains attributed to skills could partly reflect reward-shaping differences. The exact reward used for every method must be stated in the main text or a table.
  3. [§4.3, Table 3] The ablations are single numbers with no variance. The key attribution claims—that 'both the hierarchical skill design and the co-evolution mechanism are critical'—depend on deltas of 3.1–4.0 points on one benchmark. For example, Frozen Skills (28.9) vs Full Model (32.0) is a 3.1-point gap; with no error bars, this could be noise. Please provide repeated runs/seeds for at least the main ablations, and clarify exactly what 'Frozen Skills' means: is the library built from the same initial rollouts and then kept fixed, with all other components identical? Only then does the comparison isolate online evolution.
  4. [§3.2–§3.3] The method depends on a large number of hyperparameters—λ_call, merge thresholds θK_sim and θH_sim, retrieval thresholds τ_sim/θ_sim, top-K and top-M, evolution window N, retention fractions ρ_wf and ρ_ex, and library capacities. No sensitivity analysis is reported. While this alone is not fatal, it matters for the claim that the framework is broadly useful, especially because some of these thresholds directly control the trade-off between adding vs merging skills. A sensitivity study for the most critical thresholds (e.g., λ_call and the merge thresholds) would substantially strengthen the paper.
minor comments (6)
  1. [§3.2.2] Typo: 'Consitent' should be 'Consistent'.
  2. [§4.4] Typo: 'valitaion set' should be 'validation set'.
  3. [Figure 3] The tool name 'Imags.crop' appears to be a typo for 'Images.crop'.
  4. [§4.4, Figure 5] The label '~12% Faster Converge' is unclear: does it mean 12% fewer steps to reach a given success rate, or a 12% higher success rate at a fixed step? Please clarify.
  5. [§4.1.1] RL data filtering from 34,795 to ~12K samples is described at a high level. Please provide the per-step counts or a short description of each filtering stage so the reader can assess potential data-selection bias.
  6. [General] Several implementation details are deferred to the supplementary material, which is not included in the submission. For a self-contained review, please include the supplementary or at least specify baseline reward definitions and retrieval details in an appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are empirical external-benchmark comparisons with controlled ablations; the noted weaknesses are statistical-evidence concerns, not circularity.

full rationale

SPyCE makes no derivation that reduces a claimed result to its inputs. The headline result ('SPyCE consistently outperforms both RL-based and memory-based baselines') is an empirical comparison on public benchmarks (V*, HRBench, ChartQAPro, MathVerse, MathVision, WeMath, TIR-Bench); no quantity in Tables 1–2 is defined as the fitted value of another reported quantity. The 'co-evolution is critical' claim rests on the Frozen Skills ablation (Table 3, 32.0 → 28.9), which is a controlled comparison that holds the skill library fixed while keeping the reward and the training recipe unchanged; whatever its confounds (single run, no error bars), it is not circular by construction. Skill distillation uses MLLM_dk (Qwen3-VL-235B), but the paper does not define the benchmark outcome in terms of that teacher; the teacher's output is an input to the framework, not a predicted result. There are no author self-citations, no imported uniqueness theorem, and no ansatz smuggled through a citation chain. The absence of seeds/significance tests and the possibility that some gains are within run-to-run noise are statistical-evidence concerns, not circularity concerns.

Assumptions & free parameters 6 free parameters · 5 assumptions · 3 invented entities

The central claim rests on ~6 tunable/selected quantities (penalty λ, merge/retrieval thresholds, top-K, window, retention fractions, capacities) plus the unvalidated assumption that teacher-distilled skills are faithful and transferable. None of the skill-library knobs are given numeric values in the main text, so the method as printed is under-specified. The reward design (penalty on successful rollouts only) is an ad hoc choice with large ablation swings, and the training-data filter (34,795→12K) is an unquantified selection step. No new physical entities are postulated; the invented items are internal representational abstractions with no independent falsifiable handle.

free parameters (6)
  • λ_call (tool-use penalty coefficient) = 0.05
    Reward R = R_acc − λ_call·1[Succ]·N_tool (§3.1); ablation shows ±2.6–5.9 point swings depending on formulation (§4.3), indicating sensitivity.
  • merge thresholds θ_K_sim, θ_H_sim = not stated in main text
    Control whether distilled skills merge into existing entries or are added (§3.2.3); values deferred to supplementary.
  • retrieval thresholds τ_sim/θ_sim and top-K/M = not stated
    Control which execution skills are retained conditional on bottleneck descriptors and how many workflow skills are retrieved (§3.3.1).
  • evolution window N and retention fractions ρ_wf, ρ_ex = not stated
    Control how often and how much of the high-advantage rollouts are distilled back into the libraries (§3.3.2).
  • skill library capacities = not stated (Figure 4 shows a ceiling near 100 execution skills)
    Capacity triggers pruning of least-frequently-matched / lowest-success-rate entries (§3.2.3, §3.3.2).
  • RL data filtering (34,795 → ~12K samples) = ~12K
    Tool-demand screening, tool-type balancing, duplicate removal, rollout-cost control (§4.1.1); how the filter changes task distribution is not quantified.
assumptions (5)
  • domain assumption GRPO group-relative advantage is a valid training signal for skill-conditioned rollouts
    Used throughout §3.1/§3.3; GRPO's normalization assumes within-group reward comparability.
  • domain assumption Successful trajectories, summarized by a 235B teacher, yield skills that transfer across tasks and benchmarks
    Core of §3.2; never validated directly on skill quality, only through downstream accuracy.
  • domain assumption Cosine similarity in text-embedding space matches skill relevance
    All retrieval and merge decisions use cosine similarity over bottleneck/condition text (§3.2.3, §3.3.1).
  • ad hoc to paper A finished trajectory's value is captured by accuracy plus a tool-call penalty on successes only
    §3.1 reward design; ablation shows alternative formulations change results by up to 5.9 points (§4.3).
  • standard math The POMDP formalization of multimodal tool-use
    §3.1; standard formalization, no new math claimed.
invented entities (3)
  • Execution skill k = (c, u, e) — condition–action–effect triple
    purpose: Encodes reusable local visual operations (rotate, crop, zoom) with trigger condition and expected evidence gain (§3.2.1).
    Internal representational construct; only indirect evidence is the downstream ablation (removal costs 3.3 points, §4.3).
  • Workflow skill h = (b, w) — bottleneck description + workflow sketch
    purpose: Encodes high-level multi-step tool orchestration priors (§3.2.2).
    Internal construct; removal costs 4.0 points in ablation (§4.3).
  • Visual bottleneck descriptors b^(m)
    purpose: Fine-grained local obstacles generated by the policy to condition execution-skill retrieval (§3.3.1).
    Internal intermediate representation; quality never measured directly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SPyCE: Skill-Policy Co-evolution for Multimodal Agents." pith.science (2026). https://pith.science/paper/U3ZZYYWM

@misc{pith2026260713854,
  author       = {Pith},
  title        = {Pith review of: SPyCE: Skill-Policy Co-evolution for Multimodal Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U3ZZYYWM}},
  note         = {Machine review of arXiv:2607.13854}
}
read the original abstract

Multimodal agents that think with images iteratively manipulate visual evidence and invoke tools across many steps. Existing reinforcement learning methods reduce trajectories to scalar rewards, forcing the policy to discover reusable tool-use patterns from scratch on every new task; memory-based alternatives retain past experience, yet they rely on test-time retrieval, without updating the policy to absorb reusable patterns from that experience. Our key insight is that multimodal reasoning trajectories should be distilled into reusable skills that co-evolve with the policy during training, rather than being consumed as rewards or retrieved from a static store. To this end, we propose SPyCE (Skill-Policy Co-evolution), a framework that distills trajectories into a hierarchical skill library and updates it throughout reinforcement learning. Execution skills capture local visual operations, while workflow skills encode high-level priors that orchestrate tool use. During training, the policy model conditions on retrieved skills to guide its rollouts, while the skill library evolves using valuable rollouts generated by the policy. This creates a closed loop in which improved policies yield better skills, and the evolving skill library, in turn, provides stronger priors for policy rollouts. Experiments across eight benchmarks demonstrate that SPyCE consistently outperforms both RL-based and memory-based baselines. Further analysis reveals that both the hierarchical skill design and the co-evolution mechanism are critical to our design. These results suggest joint skill-policy optimization as a promising paradigm for building capable multimodal agents.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 10 linked inside Pith

  1. [1]

    Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems, 35:24824–24837, 2022

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems, 35:24824–24837, 2022. 11

    2022

  2. [2]

    Multimodal Chain-of- Thought Reasoning in Language Models.CoRR, abs/2302.00923, 2023

    Zhuosheng Zhang, Aston Zhang, Mu Li, Hai Zhao, George Karypis, and Alex Smola. Multimodal Chain-of- Thought Reasoning in Language Models.CoRR, abs/2302.00923, 2023. doi: 10.48550/arXiv.2302.00923. URL https://arxiv.org/abs/2302.00923

    arXiv doi:10.48550/arxiv.2302.00923 2023

  3. [3]

    Ddcot: Duty-distinct chain-of-thought prompting for multimodal reasoning in language models.Advances in Neural Information Processing Systems, 36: 5168–5191, 2023

    Ge Zheng, Bin Yang, Jiajin Tang, Hong-Yu Zhou, and Sibei Yang. Ddcot: Duty-distinct chain-of-thought prompting for multimodal reasoning in language models.Advances in Neural Information Processing Systems, 36: 5168–5191, 2023

    2023

  4. [4]

    Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. Visual cot: Advancing multi-modal language models with a comprehensive dataset and benchmark for chain-of-thought reasoning.Advances in Neural Information Processing Systems, 37:8612–8642, 2024

    2024

  5. [5]

    Visual chatgpt: Talking, drawing and editing with visual foundation models.arXiv preprint arXiv:2303.04671, 2023

    Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. Visual chatgpt: Talking, drawing and editing with visual foundation models.arXiv preprint arXiv:2303.04671, 2023

    arXiv 2023

  6. [6]

    Mm-react: Prompting chatgpt for multimodal reasoning and action.arXiv preprint arXiv:2303.11381, 2023

    Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Ehsan Azarnasab, Faisal Ahmed, Zicheng Liu, Ce Liu, Michael Zeng, and Lijuan Wang. Mm-react: Prompting chatgpt for multimodal reasoning and action.arXiv preprint arXiv:2303.11381, 2023

    arXiv 2023

  7. [7]

    ViperGPT: Visual Inference via Python Execution for Reasoning

    Didac Suris, Sachit Menon, and Carl Vondrick. ViperGPT: Visual Inference via Python Execution for Reasoning. pages 11888–11898, 2023

    2023

  8. [8]

    Llava-plus: Learning to use tools for creating multimodal agents

    Shilong Liu, Hao Cheng, Haotian Liu, Hao Zhang, Feng Li, Tianhe Ren, Xueyan Zou, Jianwei Yang, Hang Su, Jun Zhu, et al. Llava-plus: Learning to use tools for creating multimodal agents. InEuropean conference on computer vision, pages 126–142. Springer, 2024

    2024

Show all 36 references
  1. [9]

    Openthinkimg: Learning to think with images via visual tool reinforcement learning

    Zhaochen Su, Linjie Li, Mingyang Song, Yunzhuo Hao, Zhengyuan Yang, Jun Zhang, Guanjie Chen, Jiawei Gu, Juntao Li, Xiaoye Qu, et al. Openthinkimg: Learning to think with images via visual tool reinforcement learning. arXiv preprint arXiv:2505.08617, 2025

    2025 arXiv

  2. [10]

    Thinking with Images

    Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, and Xing Yu. DeepEyes: Incentivizing “Thinking with Images” via Reinforcement Learning.CoRR, abs/2505.14362, 2025. doi: 10.48550/arXiv.2505.14362. URLhttps://arxiv.org/abs/2505.14362

    2025 doi arXiv

  3. [11]

    VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use.CoRR, abs/2505.19255, 2025

    Mingyuan Wu, Jingcheng Yang, Jize Jiang, Meitang Li, Kaizhuo Yan, Hanchao Yu, Minjia Zhang, Chengxiang Zhai, and Klara Nahrstedt. VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use.CoRR, abs/2505.19255, 2025. doi: 10.48550/arXiv.2505.19...

    2025 doi

  4. [12]

    Thinking with programming vision: Towards a unified view for thinking with images.arXiv preprint arXiv:2512.03746, 2025

    Zirun Guo, Minjie Hong, Feng Zhang, Kai Jia, and Tao Jin. Thinking with programming vision: Towards a unified view for thinking with images.arXiv preprint arXiv:2512.03746, 2025

    2025

  5. [13]

    DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

    Zhihong Shao et al. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. CoRR, abs/2402.03300, 2024. doi: 10.48550/arXiv.2402.03300. URLhttps://arxiv.org/abs/2402.03300

    2024 doi arXiv

  6. [14]

    Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

    2023

  7. [15]

    Voyager: An Open-Ended Embodied Agent with Large Language Models.CoRR, abs/2305.16291, 2023

    Guanzhi Wang et al. Voyager: An Open-Ended Embodied Agent with Large Language Models.CoRR, abs/2305.16291, 2023. doi: 10.48550/arXiv.2305.16291. URLhttps://arxiv.org/abs/2305.16291

    2023 doi arXiv

  8. [16]

    Expel: Llm agents are experiential learners

    Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. Expel: Llm agents are experiential learners. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19632–19642, 2024

    2024

  9. [17]

    Dynamic cheatsheet: Test-time learning with adaptive memory

    Mirac Suzgun, Mert Yuksekgonul, Federico Bianchi, Dan Jurafsky, and James Zou. Dynamic cheatsheet: Test-time learning with adaptive memory. InProceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pa...

    2026

  10. [18]

    Memp: Exploring agent procedural memory.arXiv preprint arXiv:2508.06433, 2025

    Runnan Fang, Yuan Liang, Xiaobin Wang, Jialong Wu, Shuofei Qiao, Pengjun Xie, Fei Huang, Huajun Chen, and Ningyu Zhang. Memp: Exploring agent procedural memory.arXiv preprint arXiv:2508.06433, 2025

    2025 arXiv

  11. [19]

    Agent kb: Leveraging cross-domain experience for agentic problem solving.arXiv preprint arXiv:2507.06229, 2025

    Xiangru Tang, Tianrui Qin, Tianhao Peng, Ziyang Zhou, Daniel Shao, Tingting Du, Xinming Wei, Peng Xia, Fang Wu, He Zhu, et al. Agent kb: Leveraging cross-domain experience for agentic problem solving.arXiv preprint arXiv:2507.06229, 2025. 12

    2025

  12. [20]

    Instructblip: Towards general-purpose vision-language models with instruction tuning.Advances in neural information processing systems, 36:49250–49267, 2023

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale N Fung, and Steven Hoi. Instructblip: Towards general-purpose vision-language models with instruction tuning.Advances in neural information processing systems, 36:49250–49267, 2023

    2023

  13. [21]

    Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    2023

  14. [22]

    Qwen2.5-vl technical report, 2025

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Han...

    2025 arXiv

  15. [23]

    GPT-4V(ision) System Card

    OpenAI. GPT-4V(ision) System Card. Technical report, OpenAI, September 2023. URLhttps://cdn.openai. com/papers/GPTV_System_Card.pdf. Official system card, September 25, 2023

    2023

  16. [24]

    Introducing OpenAI o3 and o4-mini, 2025

    OpenAI. Introducing OpenAI o3 and o4-mini, 2025. URL https://openai.com/index/ introducing-o3-and-o4-mini/. OpenAI product release, April 16, 2025

    2025

  17. [25]

    Generative agents: Interactive simulacra of human behavior

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. Generative agents: Interactive simulacra of human behavior. pages 1–22, 2023

    2023

  18. [26]

    Planning and acting in partially observable stochastic domains.Artificial intelligence, 101(1-2):99–134, 1998

    Leslie Pack Kaelbling, Michael L Littman, and Anthony R Cassandra. Planning and acting in partially observable stochastic domains.Artificial intelligence, 101(1-2):99–134, 1998

    1998

  19. [27]

    Back to basics: Revisiting reinforce-style optimization for learning from human feedback in llms

    Arash Ahmadian, Chris Cremer, Matthias Gallé, Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ahmet Üstün, and Sara Hooker. Back to basics: Revisiting reinforce-style optimization for learning from human feedback in llms. InProceedings of the 62nd Annual Meeting of the Assoc...

    2024

  20. [28]

    Tir-bench: A comprehensive benchmark for agentic thinking-with-images reasoning.arXiv preprint arXiv:2511.01833, 2025

    Ming Li, Jike Zhong, Shitian Zhao, Haoquan Zhang, Shaoheng Lin, Yuxiang Lai, Chen Wei, Konstantinos Psounis, and Kaipeng Zhang. Tir-bench: A comprehensive benchmark for agentic thinking-with-images reasoning.arXiv preprint arXiv:2511.01833, 2025

    2025

  21. [29]

    Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? InEuropean Conference on Computer Vision, pages 169–186

    Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Yu Qiao, et al. Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? InEuropean Conference on Computer Vision, pages 169–186. Springer, 2024

    2024

  22. [30]

    Measuring multimodal mathematical reasoning with math-vision dataset.Advances in Neural Information Processing Systems, 37:95095–95169, 2024

    Ke Wang, Junting Pan, Weikang Shi, Zimu Lu, Houxing Ren, Aojun Zhou, Mingjie Zhan, and Hongsheng Li. Measuring multimodal mathematical reasoning with math-vision dataset.Advances in Neural Information Processing Systems, 37:95095–95169, 2024

    2024

  23. [31]

    Runqi Qiao, Qiuna Tan, Guanting Dong, MinhuiWu MinhuiWu, Chong Sun, Xiaoshuai Song, Jiapeng Wang, Zhuoma Gongque, Shanglin Lei, Yifan Zhang, et al. We-math: Does your large multimodal model achieve human- like mathematical reasoning? InProceedings of the 63rd Annual Meeting of...

    2025

  24. [32]

    Chartqapro: A more diverse and challenging benchmark for chart question answering

    Ahmed Masry, Mohammed Saidul Islam, Mahir Ahmed, Aayush Bajaj, Firoz Kabir, Aaryaman Kartha, Md Tah- mid Rahman Laskar, Mizanur Rahman, Shadikur Rahman, Mehrad Shahmohammadi, et al. Chartqapro: A more diverse and challenging benchmark for chart question answering. InFindings o...

    2025

  25. [33]

    V*: Guided visual search as a core mechanism in multimodal llms

    Penghao Wu and Saining Xie. V*: Guided visual search as a core mechanism in multimodal llms. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13084–13094, 2024

    2024

  26. [34]

    Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models

    Wenbin Wang, Liang Ding, Minyan Zeng, Xiabin Zhou, Li Shen, Yong Luo, Wei Yu, and Dacheng Tao. Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models. InProceedings of the AAAI Conference on Artificial In...

    2025

  27. [35]

    Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025

    Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, et al. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025

    2025 arXiv

  28. [36]

    New Embedding Models and API Updates, 2024

    OpenAI. New Embedding Models and API Updates, 2024. URL https://openai.com/index/ new-embedding-models-and-api-updates/. OpenAI product update, January 25, 2024. 13

    2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.