Pith. sign in

REVIEW 5 major objections 6 minor 39 references

Can Large Language Models Effectively Process and Execute Financial Trading Instructions?

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read No tested LLM can reliably turn natural-language trade orders into executable JSON; accuracies run 5% to 10%.

desk verdict Useful task, plausible qualitative pattern, but the reported accuracy numbers are internally inconsistent and the unreleased human labels make them unverifiable. read the letter →

arxiv 2412.04856 v1 pith:DTVOJGHH submitted 2024-12-06 cs.CE

Yu Kang , Ge Wang , Xin Yang , Yuda Wang , Mingwen Liu This is my paper · ORCID
classification cs.CE
keywords largelanguagemodelstradeorderrecognitionfinancialtradinginstructionsJSONgenerationautomatedLLMevaluationmissinginformationdetectionnaturalprocessinginfinance
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

This paper asks whether current large language models can convert a human trader's natural-language instruction into a structured, executable order. It tests five models on a 500-item Chinese-English dataset of trading instructions and finds that the models almost always produce well-formed JSON (87.50% to 98.33%) and always ask follow-up questions, yet their outputs are correct only 5% to 10% of the time. Required fields are omitted at rates from 14.29% to 67.29%, and extra, unnecessary follow-up questions appear in 76.27% to 91.53% of cases. If this holds, LLMs are not yet safe as direct parsers for trade execution without a validation layer or manual checks.

What carries the argument

The system's core is a trade order recognition pipeline plus a six-metric evaluation stack. The pipeline maps a natural-language instruction onto a fixed JSON schema with fields for strategy, symbol, order type, price, and quantity, then parses, validates, and asks for missing information before execution. The metrics are ordered from shallow to deep: Generation Rate (valid JSON produced), Missing Rate, Error Rate, Accuracy, Follow-up Rate, and Missed/Extra Follow-up Rate. The load-bearing measurement device is the 472-item manual alignment of gold JSON labels used as the reference for accuracy.

What would settle it

Take the same 500 instructions, have two independent finance annotators label executable JSON for every item, and compare the five models under exact field-level matching; if accuracy climbs well above 10% or the manual labels disagree with each other on more than a small fraction, the paper's central numbers would not survive.

Watch

Extended reading notes

Core claim

The central discovery is an empirical mismatch: five state-of-the-art LLMs (GPT-4o, GPT-4o-mini, Qwen-max-0428, DeepSeek-v2.5, and Yi-large) can generate JSON in the requested format and reliably initiate follow-up questions, but they rarely produce complete and correct executable orders. On a dataset of 500 trading instructions augmented with noise and slicing, with 472 manually aligned JSON labels as the reference, the best accuracy is 10% (Yi-large and GPT-4o) and the worst is 5% (GPT-4o-mini). Missing rates range from 14.29% to 67.29%, and the models' perfect 100% follow-up rates occur alongside missed follow-ups of 15.25% to 43.22% and extra-field rates of 76.27% to 91.53%. The paper also documents specific error modes: confusion between market and limit orders, confusion between the string 'None' or 'NULL' and true null in JSON, and unnecessary follow-up questions when the instruction is already complete.

Load-bearing premise

The conclusion rests on the 472 manually created JSON labels being a correct gold standard and on the Table 4 metric definitions accurately measuring what they name; Section 3.3 does not report inter-annotator agreement, so if the labels or metric interpretations are loose, the accuracy and missing-rate numbers shift.

Editorial extensions

If this is right

  • No tested model can be dropped directly into a live trading system as a natural-language order parser; the paper argues that manual correction or a validation pipeline is required.
  • A 100% follow-up rate is not evidence of good inquiry: every model combined it with 15.25% to 43.22% missed follow-ups and 76.27% to 91.53% extra questions, so follow-up behavior should be judged by precision, not just presence.
  • Because over-interrogation extracts additional personal and financial information, the models' tendency to over-ask carries information-security consequences for voice or chat trading interfaces.
  • The specific failure modes, such as market-versus-limit-order confusion and JSON string-versus-null confusion, identify concrete targets for prompt engineering or fine-tuning.

Reading between the lines

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

  • A natural next step, not in the paper, would be to measure how much the pipeline's validation layer actually repairs the 90% to 95% error rate; the paper demonstrates the pipeline but does not report an end-to-end accuracy after its corrections.
  • The results suggest a testable extension: fine-tune a general-purpose LLM on a few thousand noisy trade instructions and re-run the same six metrics to see whether missing and extra-field rates drop.
  • The manual gold labels are the pivot point; re-labelling with two independent annotators and reporting agreement would tell whether the 5% to 10% accuracy range is a property of the models or partly of the labels.
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

5 major / 6 minor

Summary. The paper proposes an intelligent pipeline that converts natural-language financial trading instructions into a standardized JSON format, describes a 500-item dataset generated with GPT-4o and manual refinement, and evaluates five LLMs (Yi-large, DeepSeek-v2.5, GPT-4o, GPT-4o-mini, Qwen-max-0428) using a hierarchy of metrics: Generation Rate, Missing Rate, Error Rate, Accuracy, Follow-up Rate, Missed Follow-up Rate, and Extra Follow-up Rate. The claimed findings are that models generate JSON at high rates (87.50–98.33%) and always follow up, but accuracy is only 5–10%, missing rates range from 14.29% to 67.29%, and models tend to over-inquire. The paper also describes a trading execution pipeline and a user interaction example.

Significance. The study addresses a practical and under-explored task: translating noisy natural-language trading orders into executable structured commands. If the reported numbers are validated, the result would be a useful caution about using LLMs as-is in financial execution systems, and the over-inquiry finding is a concrete behavioral insight. The paper's strength is its attempt to define an ordered set of metrics that separates generation from correctness and completeness. However, the central quantitative claims are not currently verifiable: the gold-standard labels are not released, no inter-annotator agreement is reported, and Tables 5 and 6 are internally inconsistent with the metric definitions in Table 4. No code, prompts, or raw counts are provided, so the headline 5–10% accuracy figure cannot be reproduced or checked by readers.

major comments (5)
  1. [Table 4] The note under Table 4 assigns '#' (higher is better) to Missing Rate, Error Rate, Missed Follow-up Rate, and Extra Follow-up Rate, and '*' (lower is better) to Follow-up Rate. This is reversed for all five of these metrics: higher missing/error rates are worse, missed follow-ups and extra follow-ups are worse, and a higher follow-up rate is better. The inversion contradicts the interpretation in §6.1, where high Missing Rate is explicitly treated as a defect. The symbol key must be corrected before any of the reported rates can be interpreted.
  2. [§6.1, Table 5] Table 5 cannot be reconciled with the formulas in Table 4. For GPT-4o, Generation Rate 98.33% implies about 492 JSON outputs out of 500; Missing Rate 63.56% would then mean about 313 outputs with missing fields, leaving at most 179 non-missing outputs. Correctness is also 63.56%, which would require Error Rate to be zero, but Accuracy is 10%, i.e., 50 correct outputs out of 500. For Yi-large, Generation Rate 87.50% gives 437.5 outputs; Missing Rate 14.29% gives about 62.5 missing; Correctness 82.86% gives about 362.5 non-error outputs; yet Accuracy 10% gives 50 correct outputs. The stated definitions of 'Correctness' and 'Accuracy' make these numbers mutually inconsistent. Please report raw counts and define each denominator; otherwise the headline 5–10% accuracy figure is not interpretable.
  3. [§6.2, Table 6] Table 6 reports Follow-up Rate 100% for all models while also reporting Missed Follow-up Rate values between 15.25% and 43.22%. Under the Table 4 definitions, Missed Follow-up Rate = Missing Follow-ups / Total Required Follow-ups and Follow-up Rate = Follow-ups / Total Required Follow-ups; if any required follow-up is missed, Follow-ups cannot equal Total Required Follow-ups. A rate of 100% is possible only if 'Follow-up Rate' means 'the model asked at least one question per incomplete case' rather than 'asked all required questions.' This distinction changes the abstract's claim of 'perfect follow-up rates' and must be clarified with explicit numerators and denominators.
  4. [§3.3] The gold standard for the accuracy evaluation is described in one sentence: the authors 'manually generated JSON outputs for 472 out of 500' items. No annotation protocol, per-field correctness definition, inter-annotator agreement, or release of the labels is provided. Because accuracy (5–10%) is the central quantitative result, the paper should release the dataset with labels and prompts and report agreement on at least a subset. Without this, the headline accuracy cannot be independently verified.
  5. [§5.1] Section 5.1 states that the inquiry-capability evaluation used 'a dataset of 118 trading strategy descriptions,' which is inconsistent with the paper's description of a 500-item dataset evaluated across all metrics. It is unclear whether the 118-item set is a subset of the 500, a separate set, or the base from which the 500 were derived; no cross-reference is given. The denominators of the Table 6 rates are therefore undefined, and the reader cannot determine what population the follow-up results describe.
minor comments (6)
  1. [Abstract and §6.2] The abstract's 'perfect follow-up rates' overstates the results given the 15.25–43.22% Missed Follow-up Rates in Table 6; this should be reworded once the metric definitions are clarified.
  2. [§1] The organization paragraph lists Section 6 for results but omits Section 5 (Experiment) from the roadmap, even though Section 5 contains the experimental setup.
  3. [Throughout] There are numerous typographical errors, including 'Developd' (Section 1), 'difficultly' (Section 6.2), 'mis-follow-up' (Table 7), and 'honoring of information' (Section 1); these should be corrected in a careful proofreading pass.
  4. [§6.2, Table 7, Case 2] The example labels a price question as a correct follow-up for a market order, but Section 3.3 states that market orders have price 'None' by definition; the criterion for when a price field is missing should be stated explicitly so that the labeling is consistent.
  5. [§7] The statement that 'all results are validated in our system' is not supported by any quantitative validation evidence in the paper; please either provide the validation results or soften the claim.
  6. [References] Several references are incomplete or inconsistently formatted (e.g., entries missing venue, volume, or page numbers), and some in-text mentions such as the 'Financial NLP Benchmark' are not cited; please harmonize the bibliography.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the reported metrics are empirical measurements against an author-built gold standard, and the observed metric inconsistencies are correctness and reproducibility issues rather than circular reasoning.

full rationale

Walking the claimed derivation chain, the paper performs no formal derivation, fits no parameters, and does not define any headline quantity in terms of another quantity it is supposed to predict. The central results—generation rates of 87.50% to 98.33%, accuracies of 5% to 10%, and missing rates of 14.29% to 67.29%—are empirical measurements of LLM outputs against manually constructed labels described in Section 3.3. Those labels are created from the 500 input instructions and are not defined in terms of the model outputs, so accuracy is not forced by construction. The closest potential concern is that the dataset was generated using GPT-4o while GPT-4o is also among the evaluated models (Sections 3.1 and 4.1), but the paper never claims to derive GPT-4o's performance from the dataset-generation procedure, and no equation or definition makes the evaluation equivalent to the dataset construction; this is a methodological bias concern rather than circularity. The metric formulas in Table 4 are internally inconsistent with the numbers in Tables 5 and 6—for example, GPT-4o's Missing Rate equals its Correctness, and a 100% follow-up rate coexists with 15.25% to 43.22% missed follow-up rates—but arithmetic inconsistency is a correctness and reproducibility problem, not evidence that a prediction reduces to its inputs. There are also no load-bearing self-citations: references [34,35] support only the choice to generate data with GPT-4o, and no uniqueness theorem from the authors is invoked to rule out alternatives. Consequently, no specific circular step can be quoted and exhibited, and the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No fitted numbers appear in the paper. The empirical percentages are measurements, not fitted parameters. Design choices such as 500 examples, ten-word slicing, and the five-field JSON schema are author decisions but are not free parameters in the usual fitted sense. The pipeline is a software workflow, not an invented physical or conceptual entity.

assumptions (3)
  • domain assumption GPT-4o-generated, manually refined examples simulate real-world trading conversations.
    Section 3.1 states the dataset 'simulate a variety of everyday conversations in a trading environment' and was generated using GPT-4o, so the representativeness of the benchmark rests on this premise.
  • domain assumption Manual JSON alignment for 472 of 500 items is a correct gold standard.
    Section 3.3 says manual alignment was performed, but no inter-annotator agreement, adjudication procedure, or handling of the 28 unaligned items is reported.
  • domain assumption The five selected models were run under equivalent prompts and settings.
    Section 5.1 gives the hardware and API access but not the exact system prompts, temperature, decoding parameters, or number of runs per model, so cross-model comparisons assume hidden equivalences.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can Large Language Models Effectively Process and Execute Financial Trading Instructions?." pith.science (2026). https://pith.science/paper/DTVOJGHH

@misc{pith2026241204856,
  author       = {Pith},
  title        = {Pith review of: Can Large Language Models Effectively Process and Execute Financial Trading Instructions?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DTVOJGHH}},
  note         = {Machine review of arXiv:2412.04856}
}
read the original abstract

The development of Large Language Models (LLMs) has created transformative opportunities for the financial industry, especially in the area of financial trading. However, how to integrate LLMs with trading systems has become a challenge. To address this problem, we propose an intelligent trade order recognition pipeline that enables the conversion of trade orders into a standard format in trade execution. The system improves the ability of human traders to interact with trading platforms while addressing the problem of misinformation acquisition in trade execution. In addition, we have created a trade order dataset of 500 pieces of data to simulate real-world trading scenarios. Moreover, we designed several metrics to provide a comprehensive assessment of dataset reliability and the generative power of big models in finance by experimenting with five state-of-the-art LLMs on our dataset. The results indicate that while LLMs demonstrate high generation rates (87.50% to 98.33%) and perfect follow-up rates, they face significant challenges in accuracy (5% to 10%) and completeness, with high missing rates (14.29% to 67.29%). In addition, LLMs tend to over-interrogate, suggesting that large models tend to collect more information, carrying certain challenges for information security.

Figures

Figures reproduced from arXiv: 2412.04856 by the authors.

Figure 1
Figure 1. Format Generation Performance Across Models [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Inquiry Capability Performance Across Models [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Execution Pipeline In order to address some of the mentioned issues above (e.g., incorrect JSON-generated content), we developed a comprehensive transaction execution process. This ensures that our transactions are fully executed, comprising everything from user input and parsing to the execution during the pipeline [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: User Interaction The system( [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 29 canonical work pages

  1. [1]

    Wealth Guide: A Sophisticated Language Model Solution for Financial Trading Decisions

    Das S, Lyngkhoi REZM, Saha S, and Maurya A. Wealth Guide: A Sophisticated Language Model Solution for Financial Trading Decisions. ACL Anthology 2024

    work page 2024

  2. [2]

    Revolutionizing Finance with LLMs: An Overview of Applications and Insights

    Liu T, Zhao H, Liu Z, et al. Revolutionizing Finance with LLMs: An Overview of Applications and Insights. arXiv 2024

    work page 2024

  3. [3]

    FinRL: Deep Reinforcement Learning Framework to Automate Trading in Quantitative Finance

    Liu XY, Yang H, Gao J, and Wang CD. FinRL: Deep Reinforcement Learning Framework to Automate Trading in Quantitative Finance. arXiv.org 2023

    work page 2023

  4. [4]

    Flames: Benchmarking Value Alignment of LLMs in Chinese

    Huang K, Liu X, Guo Q, et al. Flames: Benchmarking Value Alignment of LLMs in Chinese. ACL Anthology 2024

    work page 2024

  5. [5]

    Pattern recognition

    Theodoridis S and Koutroumbas K. Pattern recognition. Elsevier, 2006

    work page 2006

  6. [6]

    Large language models present new questions for decision support

    Hackathorn R, Larsen K, and Handler AK. Large language models present new questions for decision support. Decision Support Systems 2023

    work page 2023

  7. [7]

    Effective Usage of Large Language Models in Investment with an Example of Asset Allocation Analysis

    Shang K. Effective Usage of Large Language Models in Investment with an Example of Asset Allocation Analysis. 2024

    work page 2024

  8. [8]

    Quantified Strategies 2024

    Systematic Trading - What is it? - Quantified Strategies. Quantified Strategies 2024

    work page 2024

Show all 39 references
  1. [9]

    A Comparative Study of Efficiency, Performance, and Adversarial Robustness

    Fan X and Tao C. A Comparative Study of Efficiency, Performance, and Adversarial Robustness. arXiv preprint arXiv:2408.04585 2024

    2024 arXiv

  2. [10]

    Financial Sentiment Analysis(FSA): A Survey

    Man X, Luo T, and Lin J. Financial Sentiment Analysis(FSA): A Survey. IEEE Xplore 2019

    2019

  3. [11]

    News-based intelligent prediction of financial markets using text mining and machine learning: A systematic literature review

    Ashtiani MN and Raahemi B. News-based intelligent prediction of financial markets using text mining and machine learning: A systematic literature review. ScienceDirect 2023

    2023

  4. [12]

    NLP techniques for automating responses to customer queries: a systematic review

    Olujimi PA and Ade-Ibijola A. NLP techniques for automating responses to customer queries: a systematic review. Springer 2023

    2023

  5. [13]

    An Effective TF-IDF Model to Improve the Text Classification Performance

    Jain S, Jain SK, and Vasal S. An Effective TF-IDF Model to Improve the Text Classification Performance. IEEE Xplore 2023

    2023

  6. [14]

    Machine Learning Algorithms for Prediction of Stock Market: A Systematic Literature Review

    Raju SS, Teja B, Bhuvaneswari R, and Unnathi Stella G. Machine Learning Algorithms for Prediction of Stock Market: A Systematic Literature Review. Tech. rep. IEEE Xplore, 2023. url: https://ieeexplore.ieee.org/abstract/document/10486502

    2023

  7. [15]

    Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) network

    Sherstinsky A. Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) network. Neural Networks 2020;112:237–58

    2020

  8. [16]

    Long Short-Term Memory

    Hochreiter S and Schmidhuber J. Long Short-Term Memory. Neural Computation 1997;9:1735–80

    1997

  9. [17]

    GPT-FinRE: In-context Learning for Financial Relation Extraction using Large Language Models

    Rajpoot P and Parikh A. GPT-FinRE: In-context Learning for Financial Relation Extraction using Large Language Models. arXiv preprint arXiv:2303.07205 2023

    2023 arXiv

  10. [18]

    Improving stock market prediction accuracy using sentiment and technical analysis

    Agrawal S, Kumar N, Rathee G, Kerrache CA, Calafate CT, and Bilal M. Improving stock market prediction accuracy using sentiment and technical analysis. Springer 2024. 18

    2024

  11. [19]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin J, Chang MW, Lee K, and Toutanova K. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv preprint arXiv:1810.04805 2018

    2018 arXiv

  12. [20]

    Improving Language Understanding by Generative Pre-Training

    Radford A, Narasimhan K, Salimans T, and Sutskever I. Improving Language Understanding by Generative Pre-Training. OpenAI Blog 2018

    2018

  13. [21]

    Text mining arXiv: a look through quantitative finance papers

    Bianchi ML. Text mining arXiv: a look through quantitative finance papers. arXiv - QuantFin - General Finance 2024;1:1–20

    2024

  14. [22]

    Strategic behavior of large language models and the role of game structure versus contextual framing

    Lor` e N and Heydari B. Strategic behavior of large language models and the role of game structure versus contextual framing. Nature 2024

    2024

  15. [23]

    Stock price prediction using BERT and GAN

    Sonkiya P, Bajpai V, and Bansal A. Stock price prediction using BERT and GAN. arXiv.org 2023

    2023

  16. [24]

    Large Language Models in Finance: A Survey

    Li Y, Wang S, Ding H, and Chen H. Large Language Models in Finance: A Survey. arXiv 2024;1. arXiv: 2311.10723:1–20

    2024 arXiv

  17. [25]

    Deep Reinforcement Learning in Quantitative Algorithmic Trading: A Review

    Pricope TV. Deep Reinforcement Learning in Quantitative Algorithmic Trading: A Review. arXiv 2024

    2024

  18. [26]

    Deep Learning for Financial Applications : A Survey

    Ozbayoglu AM, Gudelek MU, and Sezer OB. Deep Learning for Financial Applications : A Survey. arXiv.org 2020

    2020

  19. [27]

    Overview - QuantConnect.com

    QuantConnect. Overview - QuantConnect.com. 2024. url: https://www.quantconnect.com/docs/v2/writing-algorithms/algorithm- framework/overview

    2024

  20. [28]

    A Real Time Stock tendency prognostication using Quantopian

    Rajesh P, Alam M, Tahernezhadi M, Vamshikrishna Reddy K, and Chandrakala K. A Real Time Stock tendency prognostication using Quantopian. In: 2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA). 2020:1260–7. doi: 10.1109/ICMLA51294.2020.00198

    2020

  21. [29]

    Artificial intelligence techniques in financial trading: A systematic literature review

    Dakalbab FM, Talib MA, Nasir Q, and Saroufil T. Artificial intelligence techniques in financial trading: A systematic literature review. Journal of King Saud University - Computer and Information Sciences 2024;36:102015

    2024

  22. [30]

    Quantitative Trading: An Introduction

    Liu P. Quantitative Trading: An Introduction. In: Springer, 2023. Chap. 1:1–20. url: https://link.springer.com/chapter/10.1007/978-1-4842-9675-2_1

    2023 doi

  23. [31]

    An Automated Portfolio Trading System with Feature Preprocessing and Recurrent Reinforcement Learning

    Li L. An Automated Portfolio Trading System with Feature Preprocessing and Recurrent Reinforcement Learning. In: 2nd ACM International Conference on AI in Finance (ICAIF’21). ACM. 2021:1–8

    2021

  24. [32]

    Multi-agent platform to support trading decisions in the FOREX market

    Korczak J, Hernes M, and Bac M. Multi-agent platform to support trading decisions in the FOREX market. International Journal of Production Economics 2024;228:108–22

    2024

  25. [33]

    Practical Application of Deep Reinforcement Learning to Optimal Trade Execution

    Byun WJ, Choi B, Kim S, and Jo J. Practical Application of Deep Reinforcement Learning to Optimal Trade Execution. mdpi.com 2024

    2024

  26. [34]

    Instructiongpt-4: A 200-instruction paradigm for fine-tuning minigpt-4

    Wei L, Jiang Z, Huang W, and Sun L. Instructiongpt-4: A 200-instruction paradigm for fine-tuning minigpt-4. arXiv preprint arXiv:2308.12067 2023. 19

    2023 arXiv

  27. [35]

    Self-instruct: Aligning language models with self-generated instructions

    Wang Y, Kordi Y, Mishra S, et al. Self-instruct: Aligning language models with self-generated instructions. arXiv preprint arXiv:2212.10560 2022

    2022 arXiv

  28. [36]

    GPT-4o mini: advancing cost-efficient intelligence

    Openai. GPT-4o mini: advancing cost-efficient intelligence. 2024. url: https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

    2024

  29. [37]

    Notes on Qwen-Max-0428

    Team Q. Notes on Qwen-Max-0428. 2024. url: https://qwenlm.github.io/blog/qwen-max-0428/

    2024

  30. [38]

    Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model

    Liu A, Feng B, Wang B, et al. Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model. arXiv preprint arXiv:2405.04434 2024

    2024 arXiv

  31. [39]

    Yi: Open foundation models by 01

    Young A, Chen B, Li C, et al. Yi: Open foundation models by 01. ai. arXiv preprint arXiv:2403.04652 2024. 20

    2024 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.