WhitzardNUWA Lab ↗
NUWANUWA INDEX

Search NUWA Index

Research · Benchmarks · Findings · Perspectives

中文
← NUWA Index

Evaluation dossier

CharXiv

Can models interpret research charts?

Sources and provenance ↓ Read the research narrative →

CROSS-REPORT MONITORING

PUBLISHED EVIDENCE

CharXiv · Verbalized evaluation awareness

Published measurements across source reports. Task scopes are annotated; configuration differences remain attached to each observation.

4 observations · 2 dated

● Matched◇ Configuration differs□ Some conditions undisclosed
Verbalized evaluation awareness (%) Kimi K3 DeepSeek V4 Flash 0 20 40 60 80 100 33.7% 19.3% CharXiv grounding

2 observations lack a verified model release date. They remain in Data and Systems.

Select a point or table row to inspect its conditions and provenance.

4 observations

Measurements and source conditions
System / modelResultCondition statusTask / conditionsEvaluator / reportDates
Kimi K333.7%Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease 2026-07-16Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073119.3%Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
Kimi K336.5%Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv no grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease 2026-07-16Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073110.6%Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv no grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11

Task and measurement

Benchmark protocol profile

Task / subject
The model or agent in the stated task configuration.
Measurement target
Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input.
Scoring / outcome
Attempted answer with image withheld · Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input.
Interpretation boundary
The original visual capability benchmark and the missing-image behavioral test measure different constructs.

Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input.

A reported evaluation

The source evaluates Kimi K3 · CharXiv with images withheld.

Reported measurement: Attempted answer with image withheld · 88.7%.

Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input. Withheld-image CharXiv questions; attempts include correct and incorrect guesses.

Source-reported measurements in §1.1.1; Figure 1. Transcription is not independent experimental replication.

Measurement dossier · NUWA reading

Source-method reading and retained catalogue information are distinguished. Publication does not establish independent evidence or bilingual editorial review.

Measurement construct
How well do multimodal models understand naturally occurring scientific charts?
Task population
2,323 expert-curated arXiv charts support over 10,000 descriptive/reasoning questions, split into 1,000 validation charts and a held-answer test set.
Scoring and failure modes
Separate descriptive/reasoning accuracy; validation and held-answer test populations differ.
System conditions
2,323 charts; 1,000 validation charts and remainder for test.
Statistical meaning
Source-reported result statements; this does not establish trial counts, denominators or an estimator: The paper reports GPT-4o reasoning accuracy of 47.1%. Human reasoning/descriptive accuracy was 80.5%/92.1%, and the authors flag grading or participant error as possible contributors.
Validity and interpretation
Human errors and automated grading uncertainty limit ceiling claims; chart curation is expensive and scientific-paper coverage is selective.
Appropriate research use
Model recommendation: distinguish descriptive from reasoning questions and validation from held-out test results.
  • CharXiv · AbstractSelected original resource https://arxiv.org/html/2406.18521; sha256:5d25e15e0ce1e63c2a5dca2d34c10b62b82cb45702f4071b8e870bd903cc5d30; archived source capture; exact capture time not independently established
  • CharXiv · §3 CharXiv: A Real-World and Challenging Chart Understanding BenchmarkSelected original resource https://arxiv.org/html/2406.18521; sha256:5d25e15e0ce1e63c2a5dca2d34c10b62b82cb45702f4071b8e870bd903cc5d30; archived source capture; exact capture time not independently established
  • CharXiv · §4.2 Experimental ResultsSelected original resource https://arxiv.org/html/2406.18521; sha256:5d25e15e0ce1e63c2a5dca2d34c10b62b82cb45702f4071b8e870bd903cc5d30; archived source capture; exact capture time not independently established
  • CharXiv · §5 Conclusion: LimitationsSelected original resource https://arxiv.org/html/2406.18521; sha256:5d25e15e0ce1e63c2a5dca2d34c10b62b82cb45702f4071b8e870bd903cc5d30; archived source capture; exact capture time not independently established

Evidence across reports

The same benchmark name can cover different tasks, scoring rules and system configurations. Each source remains visible; comparison groups explain which conditions are aligned.

Neo Research

Evaluating Kimi K3 for Misalignment Risks →

Actual evaluators: Neo Research

12 result records · 5 reported system names

Protocols and comparison decisions (5)
Original source charts and materials

CharXiv with images withheld

Original evaluator results · source order

Attempted answer with image withheld

0–100%
Kimi K3
88.7%
GPT-5.6 Terra
21.3%
Claude Sonnet 5
16.3%
DeepSeek V4 Flash 0731
48.2%
Showing 4 of 4 systems

Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input.

The original visual capability benchmark and the missing-image behavioral test measure different constructs. Report dates and original executor are shown with this assessment.

Evaluation conditions and system configurations
Kimi K3
Withheld-image CharXiv questions; attempts include correct and incorrect guesses. · CharXiv with images withheld
modelVersion: not specified · harness: not specified · tools: not specified · budget: not specified
GPT-5.6 Terra
Withheld-image CharXiv questions; attempts include correct and incorrect guesses. · CharXiv with images withheld
modelVersion: not specified · harness: not specified · tools: not specified · budget: not specified
Claude Sonnet 5
Withheld-image CharXiv questions; attempts include correct and incorrect guesses. · CharXiv with images withheld
modelVersion: not specified · harness: not specified · tools: not specified · budget: not specified
DeepSeek V4 Flash 0731
Withheld-image CharXiv questions; attempts include correct and incorrect guesses. · CharXiv with images withheld
modelVersion: not specified · harness: not specified · tools: not specified · budget: not specified
CharXiv with images withheld · source order
System / configurationAttempted answer with image withheldPublication / verification
Kimi K3Harness not specified88.7%Published2026-09-11
GPT-5.6 TerraHarness not specified21.3%Published2026-09-11
Claude Sonnet 5Harness not specified16.3%Published2026-09-11
DeepSeek V4 Flash 0731Harness not specified48.2%Published2026-09-11

Assessment provenance

Evaluating Kimi K3 for Misalignment Risks → · External report

Related research questions

References

  1. [1] Kimi Code — What’s New ; Kimi K3 · July 16, 2026
  2. [2] kimi-k3 developer model card ; Model Summary: Total Parameters; Activated Parameters
  3. [3] GPT-5.6 official general availability ; July 9, 2026: general availability following limited trusted-partner preview; Parameter counts not disclosed in reviewed launch article
  4. [4] Introducing Claude Sonnet 5 — Anthropic ; Publication date and availability section; Official launch article; parameter counts not disclosed
  5. [5] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; §1.1.1; Figure 1; §1.1.1 Figure 1 p.8; §1.1.3 Table 1 p.11
  6. [6] CharXiv