Pith. sign in

REVIEW 25 cited by

NoLiMa: Long-Context Evaluation Beyond Literal Matching

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2502.05167 v3 pith:6P2E2DW3 submitted 2025-02-07 cs.CL

NoLiMa: Long-Context Evaluation Beyond Literal Matching

classification cs.CL
keywords contextsmodelsneedlehaystackliterallongnolimacapabilities
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Recent large language models (LLMs) support long contexts ranging from 128K to 1M tokens. A popular method for evaluating these capabilities is the needle-in-a-haystack (NIAH) test, which involves retrieving a "needle" (relevant information) from a "haystack" (long irrelevant context). Extensions of this approach include increasing distractors, fact chaining, and in-context reasoning. However, in these benchmarks, models can exploit existing literal matches between the needle and haystack to simplify the task. To address this, we introduce NoLiMa, a benchmark extending NIAH with a carefully designed needle set, where questions and needles have minimal lexical overlap, requiring models to infer latent associations to locate the needle within the haystack. We evaluate 13 popular LLMs that claim to support contexts of at least 128K tokens. While they perform well in short contexts (<1K), performance degrades significantly as context length increases. At 32K, for instance, 11 models drop below 50% of their strong short-length baselines. Even GPT-4o, one of the top-performing exceptions, experiences a reduction from an almost-perfect baseline of 99.3% to 69.7%. Our analysis suggests these declines stem from the increased difficulty the attention mechanism faces in longer contexts when literal matches are absent, making it harder to retrieve relevant information. Even models enhanced with reasoning capabilities or CoT prompting struggle to maintain performance in long contexts. We publicly release the dataset and evaluation code at https://github.com/adobe-research/NoLiMa.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 25 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. LifeSide: Benchmarking Agents as Lifelong Digital Companions

    cs.CL 2026-06 unverdicted novelty 7.0

    LifeSide is a new benchmark that evaluates AI agents on multi-session Memory-Emotion-Environment loops via simulated user profiles and event trajectories, revealing that models saturating existing memory tests fail at...

  2. ContextEcho: A Benchmark for Persona Drift in Long Agentic-Coding Sessions

    cs.CL 2026-05 unverdicted novelty 7.0

    ContextEcho benchmark shows persona drift occurs across 23 frontier models in long agentic-coding sessions, is not reliably reset by compaction, and can be restored by single-shot anchors with mode-dependent effects.

  3. Positional Failures in Long-Context LLMs: A Blind Spot in Reasoning Benchmarks

    cs.CL 2026-05 conditional novelty 7.0

    Audits reveal no reasoning benchmark controls position/filler/length jointly; CRE shows LLMs drop up to 88pp on middle-position tasks at 64K context, with diagnostic probe supporting positional cause.

  4. Retrieval from Within: An Intrinsic Capability of Attention-Based Models

    cs.LG 2026-05 unverdicted novelty 7.0

    Attention-based models can intrinsically retrieve and reuse pre-encoded evidence chunks via decoder attention queries, unifying retrieval with generation and outperforming external RAG pipelines on QA benchmarks.

  5. Reasoning over Video: Evaluating How MLLMs Extract, Integrate, and Reconstruct Spatiotemporal Evidence

    cs.CV 2026-03 unverdicted novelty 7.0

    VAEX-BENCH shows state-of-the-art MLLMs perform substantially worse on abstractive spatiotemporal reasoning tasks than on matched extractive tasks in video understanding.

  6. Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions

    cs.CL 2025-07 unverdicted novelty 7.0

    MemoryAgentBench is a new multi-turn benchmark assessing four memory competencies in LLM agents—accurate retrieval, test-time learning, long-range understanding, and selective forgetting—showing that existing methods ...

  7. HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following

    cs.AI 2026-07 conditional novelty 6.0

    A new 65-task benchmark measures whether AI agents obey long company handbooks across multi-tool workflows; the best model passes 36.2% under strict grading.

  8. Where Facts Go Missing: A Layerwise Taxonomy and Per-Layer Attribution of Information Omission in Air-Gapped LLMAgent Pipelines

    cs.MA 2026-07 conditional novelty 6.0

    In a controlled 75,476-trial stress test, about 73% of omitted-fact failures in LLM agent pipelines are traced to deterministic middleware (redaction, pagination, truncation) rather than model behavior.

  9. CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents

    cs.LG 2026-07 conditional novelty 6.0

    Decoupling file reads from file contents and re-injecting current repository state at each step cuts LLM coding-agent context use by up to 50% without hurting pass rates.

  10. UNIBROWSE: A Data-to-Agent Framework for Multimodal BrowseComp

    cs.CL 2026-07 conditional novelty 6.0

    A unified KG-plus-live-web data pipeline covering all three multimodal BrowseComp information-flow patterns, plus an exploration-degree filter, yields a 35B agent at 54.4 avg accuracy.

  11. What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents

    cs.LG 2026-07 conditional novelty 6.0

    KV-cache eviction, prompt compression, recurrent state bounding, and agent memory consolidation are unified as one rate-distortion problem with a shared lower bound, shared failure mode, and transferable mechanisms.

  12. KForge: LLM-Driven Cross-Platform Kernel Generation for AI Accelerators

    cs.LG 2026-06 unverdicted novelty 6.0

    KForge uses dual LLM agents for cross-platform kernel generation, reporting 2.12% throughput gain on NVIDIA B200 vs TensorRT-LLM and 5.13x geometric mean speedup on Intel Arc B580 vs PyTorch on 37 workloads.

  13. Separating Semantic Competition from Context Length in RAG Reading

    cs.CL 2026-05 unverdicted novelty 6.0

    A matched-control protocol on SQuAD recovers 4.5-6 EM points and 0.057-0.068 F1 in two small models by swapping hard distractors for easier ones, isolating semantic competition from context length.

  14. Parallel Context Compaction for Long-Horizon LLM Agent Serving

    cs.AI 2026-05 unverdicted novelty 6.0

    Parallel compaction for LLM agent context management provides predictable volume control and reduces wall time versus sequential baselines on HotpotQA and LoCoMo.

  15. Where Does Long-Context Supervision Actually Go? Effective-Context Exposure Balancing

    cs.CL 2026-05 conditional novelty 6.0

    EXACT re-allocates training supervision by inverse frequency of long effective-context targets, improving NoLiMa and RULER scores by 5-18 points on Qwen and LLaMA models without degrading standard QA or reasoning.

  16. Slipstream: Trajectory-Grounded Compaction Validation for Long-Horizon Agents

    cs.MA 2026-05 unverdicted novelty 6.0

    Slipstream uses asynchronous compaction with trajectory-grounded judge validation to improve long-horizon agent accuracy by up to 8.8 percentage points and reduce latency by up to 39.7%.

  17. Retrieval from Within: An Intrinsic Capability of Attention-Based Models

    cs.LG 2026-05 unverdicted novelty 6.0

    Attention-based models can retrieve evidence intrinsically by using decoder attention to score and reuse their own pre-encoded chunks, outperforming separate retrieval pipelines on QA benchmarks.

  18. Positional Biases Shift as Inputs Approach Context Window Limits

    cs.CL 2025-08 conditional novelty 6.0

    The lost-in-the-middle effect in LLMs appears mainly when inputs fill up to half the model's context window; beyond that, accuracy favors information closest to the end.

  19. EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering

    cs.CL 2026-06 unverdicted novelty 5.0

    EASE-TTT creates a soft attention target from evidence chunks to guide query-side test-time adaptation, yielding higher macro-average scores than full-context, retrieval-only, and standard qTTT baselines on six LongBe...

  20. Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces

    cs.CL 2026-06 unverdicted novelty 5.0

    Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.

  21. Tool Attention Is All You Need: Dynamic Tool Gating and Lazy Schema Loading for Eliminating the MCP/Tools Tax in Scalable Agentic Workflows

    cs.AI 2026-04 unverdicted novelty 5.0

    Tool Attention cuts tool-related tokens by 95% and raises context utilization from 24% to 91% in a 120-tool simulation via dynamic gating and lazy loading.

  22. Context Collapse: Barriers to Adoption for Generative AI in Workplace Settings

    cs.CY 2026-04 unverdicted novelty 5.0

    Expert interviews demonstrate that context in generative AI workplace use collapses or rots over time, limiting tool effectiveness and revealing pitfalls in computational context approaches.

  23. MemOCR: Layout-Aware Visual Memory for Efficient Long-Horizon Reasoning

    cs.AI 2026-01 unverdicted novelty 5.0

    MemOCR renders structured memory as images with adaptive visual density to improve long-horizon reasoning under tight context budgets.

  24. MiMo-V2-Flash Technical Report

    cs.CL 2026-01 unverdicted novelty 5.0

    MiMo-V2-Flash is a 309B/15B MoE model trained on 27T tokens with hybrid attention and multi-teacher on-policy distillation that matches larger models like DeepSeek-V3.2 while enabling 2.6x faster decoding via repurpos...

  25. Not All Needles Are Found: How Fact Distribution and Don't Make It Up Prompts Shape Retrieval, Reasoning, and Hallucination in Long-Context LLMs

    cs.CL 2026-01 conditional novelty 5.0

    On a new extended needle-in-a-haystack benchmark, explicit anti-hallucination prompts and dispersed fact placement cause some long-context LLMs to over-refuse or collapse in accuracy, while others remain robust.