Benchmark saturation and data contamination increasingly obscure genuine scientific reasoning in frontier LLMs. We introduce \textsc{ScienceArena}, an olympiad-style benchmark from thirteen public science competitions in physics, chemistry, and biology, including IPhO and IChO 2025--2026, IBO 2023, USAPhO 2026, and USNCO 2025. Its open-ended, multi-step problems use process-credit rubrics, making faithful scoring difficult. We build ScienceArena through an expert-audited digitization pipeline that converts official exams, figures, solutions, and rubrics into structured items verified by olympiad medalists. To scale evaluation beyond costly human grading, we calibrate LLM-as-judge against medalist ground truth on archived answers from five models across IPhO and IChO; two strong judges stay within one point of expert total scores. Medalist notes show that failures often stem from visual grounding, structure fidelity, and global problem control rather than missing terminology. Evaluating fourteen recent LLMs with interleaved solving, we find that top models obtain medal-equivalent rubric scores on several public international exams, while chemistry and long-horizon consistency remain key bottlenecks. We provide an interactive \href{https://science-arena.onrender.com/}{demo}.
Large language models can now generate complex, multi-step mathematical proofs, but reliably determining their correctness and localizing early logical errors remains a critical challenge. Existing evaluation approaches largely depend on model-based natural-language judgments, which often overlook local reasoning gaps. While formal theorem provers like Lean offer a path to rigorous verification, using them to evaluate informal text requires solving locality and semantic mismatches: a prover might bypass a local flaw by proving an overly broad target, or validate an auto-formalized statement that drifts from the original mathematical intent. To address this, we introduce FaithSieve, a Lean-assisted framework for fine-grained evaluation of natural-language mathematical proofs. FaithSieve decomposes coarse proof steps into local reasoning units, extracts typed proof obligations, and verifies them through a formal evaluation agent. Formal validation is gated by semantic alignment scoring, so Lean evidence is incorporated only when the formal statement faithfully preserves the context, objects, and logical form of the original claim. We construct two expert-verified datasets, ProofLoc-Olympiad and ProofLoc-University, to benchmark first-error localization. On the 350-problem Olympiad dataset, FaithSieve using a GPT-5.4 backbone achieves 81.43% exact first-error accuracy, outperforming the direct-judging baseline of 72.29%. Furthermore, on the 200-problem ProofLoc-University benchmark spanning six advanced domains, FaithSieve reaches 84.5% exact accuracy, compared to 75.0% for the direct judge. Our work demonstrates that decomposing proofs into fine-grained units and grounding them with faithful formal evidence significantly improves reliable evaluation of natural-language reasoning.
As large language models are increasingly used in data-scarce and evolving task scenarios, few-shot in-context learning (ICL) has become a key paradigm for task adaptation. However, direct ICL often uses a small set of examples without explicitly abstracting task rules, making it sensitive to example construction. In contrast, human learners often reduce such sensitivity by first summarizing task rules from examples and then applying them to new instances. To evaluate this ability, we propose StrategyBench, which selects strategy-inducible tasks from BIG-Bench, constructs reference strategies, and defines evaluation metrics along two dimensions: strategy quality and downstream utility. We further analyze strategy induction from three perspectives: task variation, model configuration, and adaptation setting, covering category-wise differences, generator-executor choices, demonstration design, and SFT-based adaptation. Experiments show that explicit strategy utility differs substantially across task categories and depends on both strategy generation and execution conditions. The benchmark is released at: https://anonymous.4open.science/r/StrategyBench-D53C.
Ziyue Wang, Aomufei Yuan, Yiran Yao +10cs.CL cs.AI
Large language models are increasingly used to propose research ideas, yet the prevailing ways of judging such ideas supply no shared decision rule: free-form judging sways with style and position, and scoring against a later paper rewards recovery of one realized trajectory. We introduce a benchmark that carries a proposal from Literature to Test: the Lit2Test benchmark centers on a six-field contract organized around a falsifying outcome, so that every proposal precommits the observation that would prove it wrong, making its quality decidable in the first place rather than merely arguable. Built prospectively from 200 real-paper neighborhoods, Lit2Test elicits proposals from four frontier models and compares them through 1,200 pairwise comparisons judged blind in both presentation orders. The protocol audits its own reliability through diagnostic controls and bounded human calibration, with three annotators corroborating the conclusions within explicitly stated reliability bounds. Lit2Test recovers a strict ranking of the four models in all 10,000 bootstrap replicates, and the separation comes from the quality of the proposed tests and metrics rather than from surface fluency. We release the benchmark, construction pipeline, and audit artifacts for public use.
Formal proofs in Lean 4 that pass the kernel's type checker can nonetheless vary widely in quality. We introduce ProofJudge, an agentic LLM-as-judge system that scores formal proof quality along five dimensions beyond correctness: library leverage, automation fit, structural clarity, statement quality, and Mathlib conventions. We evaluate ProofJudge on a novel dataset of 218 declarations drawn from distinct Mathlib PRs. The judge agent is grounded by tool access to the commit the PR is applied to, enabling it to query the library state when scoring. A judge is considered aligned with human preferences when it rates the version of the PR Mathlib accepted above the initial version that was sent back for revision. All six judge models evaluated recover the reviewers' preference well above chance, from 80.8% to 63.5%, and two open-weight judges reach roughly 70% at a tenth of the best judge's cost. We release the judge harness, evaluation dataset, and evaluation traces as open-source artifacts to support further research.
Reasoning in LLMs is overwhelmingly studied in domains that provide a model with rules: mathematics and code. Linguistic puzzles invert this: the solver must first discover the system before reasoning within it. We present the IOL-AI Challenge, an open-science competition run on the unseen problems of the International Linguistics Olympiad (IOL) 2026 Individual Contest, evaluated both automatically and, for the first time, by members of the official IOL Jury under the same rubrics applied to human contestants. The challenge drew 731 submissions from 46 teams under a strict compute budget (one T4, 30 mins). We additionally benchmark 15 unconstrained frontier and open models, with Claude Opus 4.8 earning a jury score equivalent to a gold medal, while both resource-constrained systems we submitted for jury grading scored in the range of the bottom 5% of contestants. Capability was not determined by scale: 14B submissions outperform models twice their size, and gains come from decoding and output-handling rather than model capacity. We also found that automatic metrics rank systems exactly as the jury does, but compress the scale, upscoring weak systems by ~13 points and understating strong ones. Our analysis shows that while frontier models might have prior knowledge about some of the problem languages, it does not significantly help them solve the linguistic reasoning tasks, leaving linguistic reasoning as a strong benchmarking proxy for generalizable reasoning skills.
How should we assess whether large language models can perform mathematical invention? I argue that this question is currently underspecified: mathematical creativity is not one capacity but several mechanistically distinct modes of meaning-making - reflexive introspection on mathematical practice, analogical import from the sciences, problem-driven construction, and the bridging of distant domains - together with a further, cross-cutting distinction between meaning pursued because a pattern was observed and meaning pursued because it is strategically wanted, a distinction I develop through the case of conjecture-formation. These mechanisms are likely non-substitutable, so that competence in one does not transfer to the others. Grounding each in a historical case study and in an architecture-level account of current transformer-based systems, I suggest that today's models concentrate their competence in modes shaped by recombination and search over existing building blocks; if that description holds, the remaining modes are out of reach in principle, not just slower - though whether it holds is itself the open, empirical part. Because proof is getting cheaper as AI improves at generating it - a shift the field's own leading voices are now diagnosing - mathematical value is migrating toward the modes current systems cannot yet perform, and evaluations of AI mathematical ability should be organized around this taxonomy rather than around aggregate benchmarks that conflate it.
The generation of mathematically precise diagrams from tex- tual prompts has emerged as a critical yet underexplored capability of Large Language Models (LLMs). This has been of interest to researchers in the areas of curriculum preparation, automated ranking of problem sets, and scientific publishing. For LLMs to achieve this, it requires per- fect coordination between Spatial Reasoning, Mathematical Reasoning, and Rendering systems. While existing benchmarks such as MathVision, MathVista are built for Math Reasoning or DiagramGenBenchmark, Mer- maidSeqBench on general purpose diagram generation, no prior work provides a standardized set of prompt, image pairs that can be used to evaluate the LLMs specifically on math diagram generation. This includes fields that span both both text-to-code and text-to-image paradigms. We introduce Math-Vision Diagrams, the first benchmark specifically designed to evaluate LLMs on mathematical diagram generation, and the first to assess text-to-code and text-to-image generation paradigms together in a single unified setting, agnostic of the underlying coding lan- guage or model type. Building on the Math-Vision benchmark, we select a subset of 2920 images out of 3040 from high-quality competition problems with essential visual context. A novel pipeline combining an ensemble of LLMs with Subject Matter Expert (SME) curation is presented, together with a suite of evaluation metrics. Testing several leading models against this benchmark, we demonstrate that LLMs struggle with math diagram generation. All code, data, curation pipeline, and evaluation scripts will be fully open-sourced.
Can a model look at a river delta and a lightning bolt and see that they share a structure? We introduce GEB-Bench, a benchmark whose unit is an abstract structural motif--self-reference, a strange loop, a Mobius twist--in the spirit of Godel, Escher, Bach. Each motif is told in several voices: a natural scene whose composition is the structure, a folk story whose telling enacts it through a mechanically checkable form device, a mathematical theorem, and a programmatic skeleton; surface parameters are declared nuisance variables and never scored. Motifs, voices, and the structural changes between them form a small cross-modal category, and GEB-Bench's tasks are its questions. Evaluating twelve open and proprietary models, we find that abstraction failure is lawful. The central finding is a gap between recognition and cross-voice mapping: models identify a structure within one voice far better than they carry it across voices; every model pays this tax, and mapping strong enough to narrow it appears only at the frontier tier. Two patterns support it. Errors align more strongly with the designed formal geometry than with measured perceptual geometries, and frontier models from different vendors converge on the same wrong answers; and surface complexity taxes every model that reads structure, with capacity buying headroom rather than immunity. GEB-Bench is fully generative and released with its pipeline.
Large language models can solve substantially harder reasoning problems with more inference-time compute. The term "test-time scaling," however, now covers diverse inference algorithms that extend deliberation along a single trajectory, sample completed candidates and aggregate them through voting or verification, or search over unfinished partial states. These algorithms differ in their statistical structure, compute accounting, and failure modes. Treating these procedures as interchangeable under a single scalar "budget," or reporting accuracy without the inference protocol that produced it, makes results difficult to compare across studies. We develop a systematic account of test-time scaling along three axes. First, we formalize test-time scaling as budgeted inference over the implicit prefix tree of an autoregressive model and distinguish three structural regimes: single-trajectory sequential scaling, leaf-level scaling with terminal reduction, and prefix-level scaling. Second, we treat the evaluated object as the entire inference system and develop evaluation principles that separate end-to-end system performance from candidate-bank diagnostics. We introduce an evaluation profile whose coordinates and simple functionals recover or bound common repeated-sampling metrics, and prescribe protocol-matched reporting of compute and uncertainty. Third, we specify reproducibility requirements for inference protocols, distinguishing exact replay from distributional reproducibility and identifying the artifacts needed to support each. We also organize the open-weight reasoning ecosystem by model-side and interface mechanisms, apply these principles to broad-knowledge, symbolic-reasoning, and competition-mathematics benchmarks, and assemble over 2 billion full reasoning traces for release with progressively richer verifier and token-level signals.
Large language models increasingly generate optimization models from natural language, but existing evaluation often reduces a generated model and its ground truth to a single equivalent/not-equivalent verdict or an execution-success rate--labels that are neither independently checkable nor faithful to the multiple distinct senses in which two formulations can agree. We present ModelEquivBench, a certifying, multi-relational evaluation system that reports a per-pair semantic profile E0--E6: model construction and exact ingestion (E0), verified representation alignment (E1), same-space and projected feasible-set relations (E2, E3), objective-order equivalence (E4), optimal-value equality (E5), and optimizer-set equivalence (E6). Each decided entry carries relation-appropriate, independently re-checkable evidence: replayable traces or explicit maps for E0--E1, exact-rational certificates for positive E2--E6 conclusions, and explicit witnesses for supported negatives. Incomplete mapping search, unsupported structure, and resource limits produce typed UNKNOWN or N/A outcomes rather than guesses, while unmet prerequisites are reported as ABSENT. Using ModelEquivBench to evaluate three model snapshots--GPT-5.4, Claude Sonnet 4.6, and Qwen3.5-397B-A17B--on the same frozen cohort of 173 base problems (346 cells per model) under a no-repair protocol, the resulting profiles expose distinctions that coarse baselines do not represent: 49, 35, and 25 cells contain executable candidates that are nevertheless certified negative on at least one supported relation, and 25, 8, and 18 structural rejections occur on pairs for which E2 certifies mapped feasible-set equality under a verified map. The three model snapshots fail at different stages of the profile and therefore cannot be meaningfully reduced to a single accuracy score.
Diagrams are widely used to support logical reasoning, and prior studies suggest that representations such as Euler diagrams can improve human reasoning performance. Recent work has also explored their effects on large language models (LLMs). In this paper, we compare four representational conditions for syllogistic reasoning: natural language, logical notation, linear diagrams, and Euler diagrams. Using 285 problems from Ando et al. (2024), we evaluate two contemporary LLMs, Claude 3.5~Sonnet and GPT-4o-mini. Our results show that diagrammatic representations do not consistently improve performance. Although the models perform well on entailment and contradiction problems, they struggle with neutral problems and often make systematic conversion errors. Overall, the results suggest that the tested models gain limited benefit from diagrams in logical reasoning tasks.
Introducing Relay-Bench, an unsaturated, holistic, text-only benchmark that measures LLMs' ability to complete an assortment of tasks from distinct domains in a single prompt. The leading model, GPT-5.5 (xHigh), scores 43.3%. The test set entirely consists of composite problems: groups of single-domain subproblems that are strung together into challenges that require reasoning across multiple domains in combination. Many of these problems then have layers of complexity added through prompt encoding and deliberate context bloat. Domains tested include visual reasoning, coding, math, information extraction (with a focus on web search), problem-solving, general knowledge, and data analysis. No restrictions are imposed outside of the model harness, and models are explicitly encouraged to leverage code-execution, web searches, and all available tools. All problems are composed of two to thirteen subproblems and do not require multi-modal input or output.
Large language models (LLMs) frequently contradict themselves when the surface form of a logically equivalent question changes. We present a benchmark of 350 question families (1,750 total questions) for Controlled Reformulation Testing (CRTBench) to evaluate logical invariance. In this benchmark, we investigate LLMs' ability to maintain consistent answers across controlled reformulations, which include contrapositive rewriting, double negation, negation flipping, and passive voice. We evaluate several frontier LLMs and observe an accuracy-consistency gap where GPT-5.4-mini achieves $98.9\%$ base accuracy but only $60.3\%$ family-level consistency, while reasoning-optimized o4-mini achieves $96.9\%$ consistency. From our experiments, we observe that failures cluster around logically nontrivial transformations such as contrapositive rewriting ($72.4\%$ for GPT-5.4-mini) and double negation ($84.6\%$), while surface-level rephrasing remains robust ($94-100\%$). Increasing reasoning effort improves GPT-5.4-mini to $85.4\%$ consistency, but leaves GPT-5.4 unchanged overall because gains on nested negation are offset by failures on quantifier families. These results show that accuracy alone is not enough for evaluating logical reasoning in LLMs.
When LLMs exhibit uneven performance across planning tasks, these gaps are often attributed to task difficulty. We argue that this explanation is incomplete, as task-level variation may reflect distinct latent planning competencies rather than differences along a single ability spectrum. We study this question on ACPBench-Hard by evaluating multiple LLM families under varying test-time reasoning budgets and applying a multidimensional item response theory model to uncover the latent competency structure underlying LLM planning. The analysis reveals two principal dimensions that shape planning performance: operational reasoning, the ability to evaluate local action applicability and immediate state transitions, and structural enumeration, the ability to reason about goal reachability and landmark structure. Operational reasoning improving under model scaling and longer reasoning traces, while structural enumeration remains comparatively insensitive. Our findings motivate competency-level evaluation of LLM planning, shifting the focus from whether models improve overall to which planning competencies improve, under what conditions, and why.
Large language models (LLMs) are increasingly evaluated on mathematical problem solving, yet prior work often treats representationally equivalent formulations as interchangeable and conflates reasoning errors with interface failures. This paper investigates representation robustness in LLM-based mathematical problem solving by systematically varying surface representations of the same underlying problems, including story problems, word-equations, symbolic equations, and isomorphic paraphrases. Using a curated dataset of mathematically equivalent problems, we evaluate five contemporary LLMs under a direct answer generation condition. We find substantial representational sensitivity: models frequently change correctness across equivalent formulations, with nontrivial flip rates across story, symbolic, and word-equation variants. We also observe systematic regressions under isomorphic reformulations, showing that even subtle paraphrase-level changes can degrade performance despite preserved mathematical structure. We then evaluate a code-augmented condition in which models externalize reasoning as executable Python code that is run locally for validation. This interface reveals strong latent reasoning capability in some models that perform poorly under direct prompting, but it does not uniformly improve robustness. Instead, failures shift across interaction layers, from opaque reasoning errors to protocol violations and execution failures. Even when executable reasoning succeeds, representation sensitivity often persists. Overall, our results show that reasoning scaffolds do not eliminate representational brittleness, but expose new tradeoffs among correctness, reliability, latency, and cost. We argue that representation should be treated as a first-class interface design variable in LLM evaluation and deployment, especially for AI-assisted problem-solving systems.
Current large-language-model (LLM) physics benchmarks are usually scored by answer accuracy, which cannot distinguish genuine reasoning from recall of familiar problem patterns and reveals little about where a model's reasoning breaks down. We introduce an auditable four-stage diagnostic that evaluates whether an LLM can reason inside an unfamiliar physics framework through induction, formulation, prediction, and review. The diagnostic combines locked pre-registrations, fresh sessions between stages, dual-LLM judging, and a human-audit pathway, and we apply it to three parallel physics worlds: a single-equation counterfactual world ($F=mv$), a historical framework (Aristotelian mechanics), and a four-domain counterfactual world (Decay World). Across Claude Opus 4.7, GPT-5.5, and Gemini 3.1 Pro, the three worlds yield composite PASS rates are 6/15, 6/15, and 0/15 respectively (content $\land$ structural for $F=mv$ and Aristotelian, content axis only for Decay World where the structural axis is out of scope). The most pointed empirical pattern is a qualitative-versus-quantitative asymmetry: in Decay World, models almost never predict the wrong direction of change, but frequently compute the wrong ratio by slipping back to standard-physics relations. The protocol also surfaces two methodology findings: LLM-judge reliability does not transfer across frameworks, and Stage 4 self-review is weak in every framework, with the model's own review wrongly reporting no earlier error in at least two-thirds of the trials that actually contained one. We release the full prompts, responses, verdicts, and audit records.
We introduce the Complexity Ceiling Benchmark (CCB), a controlled evaluation of how language-model reasoning decays as the number of required sequential steps grows. CCB fixes the semantic content of a task and varies only its depth N in {5,...,50} across three structurally distinct regimes: grounded spatial state-tracking, abstract symbolic pointer manipulation, and transitive relational inference. Across 6,000 trials over five frontier and open-weight LLMs we find a consistent pattern of geometric per-step decay with widely separated domain ceilings: on the first two regimes the strongest models retain pd>0.92 across N=50; on the third every model collapses by N=5, with the best model's 50%-success horizon at H0.5~4.7 steps despite pd=0.863. A trace-level metric (TFBC) shows that 14.5% of correct answers across the benchmark are reached via incorrect intermediate reasoning. Forced verbose state-tracking does not move the ceiling (McNemar p=1.000), and the mean step at which reasoning first diverges, k*, predicts within-domain accuracy better than parameter count. CCB and the geometric decay model together reduce a model's long-horizon reasoning profile to one interpretable number per task family.
Large language models are increasingly deployed as investment research assistants, yet no benchmark tests whether they can accurately reconstruct and apply the specific procedural decision frameworks of expert investors. We introduce InvestPhilBench, a multi-layer dynamic benchmark spanning eight cognitive tiers, from principle identification (L1) to novel framework extrapolation (L8). The v0.6 release comprises 118 primary-source-verified investment principle cards, 25 decision framework cards with explicit topology metadata, and 243 QA questions (197 dev / 46 held-out test). For reproducible scoring at scale we introduce the Benchmark Automated Scoring Pipeline (BASP) -- five algorithmic metrics (OGRS, KCCS, SAP@k, IVP, CKCA) -- the Failure Mode Detection Protocol (FMDP) with computable rules for six failure modes, and Gate Reconstruction Accuracy (GRA), a per-gate metric for questions with gold reasoning programs. In this release, InvestPhilBench is primarily a benchmark-and-methodology contribution. A four-model sanity wave on the 188-question development split shows a sharp provider-tier split (BASP 0.906 vs. 0.438); these mixed-judge numbers are confounded upper bounds. The central finding: the BASP composite saturates at the frontier (Claude L4 = 0.932) while GRA still exposes a procedural deficit (frontier L4 GRA approx. 0.77, L7 GRA 0.57-0.62) -- composite scoring rewards fluent prose and hides the procedural gap. v0.6 implements a unified judge and true model-in-the-loop retrieval/oracle conditions; the de-confounded multi-model leaderboard and full three-condition run are v1.0 deliverables. On a 100-item expert-annotated gold set the automated BASP composite tracks the human reference at Pearson r = 0.72 (MAE = 0.10), with attribution (SAP@3) the weakest sub-metric and the failure-mode detector running sensitive-but-over-flagging.
This work investigates the ability of large language models (LLMs) to generate mathematical equations from scientific texts. Prior work faces challenges in unstructured grounding, multi-equation dependency, and humanaligned evaluation. To this end, we construct a dataset of AI research papers, pairing contextual passages with ground-truth equations and variable descriptions. We develop an explainable equation generation workflow and evaluate it across diverse open- and closed-source LLM backbones. We introduce an evaluation protocol combining automatic metrics, LLM-based rubrics, and human judgments to assess accuracy, explainability, and human-LLM alignment. Results indicate that LLMs perform moderately on lexical- and syntactic-based similarity, while struggling with semantic accuracy. Comparisons between LLM-based evaluations and human judgments reveal limited alignment, highlighting challenges in using LLMs to assess equation quality. These findings offer insights for improving equation generation models and developing more reliable evaluation methods for scientific text. We provide code and data for reproducibility.
Large language models (LLMs) are increasingly capable of mathematical problem solving and can even assist with research-level proofs, yet we still lack a scalable and reproducible way to measure step-level reasoning in long proofs across diverse sources. This evaluation gap limits trustworthy AI assistance in proof-certified scientific progress. Existing evaluations often emphasize final answers or rely on costly expert grading, while end-to-end proof generation remains open-ended and hard to verify automatically. We introduce Mask-Proof, a pipeline that turns real proofs into automatically checkable masked-step tasks. It masks key formula steps, provides the necessary surrounding context, and evaluates model reconstructions with an LLM-based equivalence judge using repeated votes for stability. The resulting Mask-ProofBench contains 292 curated problems across diverse research areas. Experiments with 17 models show that reasoning-enhanced models outperform standard models by 12% to 27%. Our evaluator achieves 96.8% agreement with expert annotators, enabling faithful, reproducible, and comparable measurement of step-level mathematical reasoning. Benchmark, annotations, and code are available at https://github.com/weating/Mask-Proof.
Math reasoning benchmarks have proliferated, yet most lack a per-item difficulty signal grounded in actual human performance. We introduce KCSAT-ML, a decade (2014-2025) of Korean College Scholastic Ability Test (KCSAT; Suneung) mathematics: 664 problems with a 339-item core set carrying official per-item error rates from nationwide cohorts of hundreds of thousands of examinees. We pair the benchmark with Difficulty-aligned Reasoning Gain (DRG): a score-orthogonal metric that asks whether a model's mistakes concentrate on the items humans found hard, or on items humans found easy. Together they expose, across a wide range of VLMs (and LLMs via OCR), three patterns: (i) low-budget accuracy collapses on the high-human-error tail at every model size; (ii) test-time scaling (TTS) raises token use roughly linearly with cohort error rate, while accuracy gains follow a non-monotonic curve; (iii) within a single family, TTS flips between anti-scaling on the hardest items and overthinking on easier ones -- two faces of the same alignment failure. On DRG, models with near-identical accuracy can sit at near-opposite values: one model gets wrong what humans also find hard, while another solves the hardest items yet fails on items humans find easy -- a contrast that aggregate accuracy hides. Our code and dataset builder will be open-sourced at https://github.com/naver-ai/KCSAT-ML.
Arslan Bisharat, Brian Ortiz, Eric Spencer +5cs.AI cs.LG cs.LO cs.SE
TLA+ has supported industrial verification at companies such as Amazon and Microsoft, yet writing correct TLA+ specifications from natural language still requires time and expertise, which limits adoption. LLMs show promise, but no prior study measures whether they produce semantically correct TLA+ specifications from natural language. This paper presents the first systematic evaluation of LLM-based TLA+ specification synthesis from natural language. Our study evaluates 30 LLMs across eight families on a curated dataset of 205 TLA+ specifications: 25 open-weight models across four prompting strategies (2,600 runs) and 5 proprietary models under few-shot prompting (130 runs), all validated by the SANY parser and TLC model checker. LLMs achieve up to 26.6% syntactic correctness but only 8.6% semantic correctness, with successes exclusive to progressive prompting. Results show that model size does not predict quality, e.g., DeepSeek r1:8b outperforms its 70B variant across all strategies, which suggests the importance of reasoning alignment for formal languages. Code-specialized models consistently underperform due to negative transfer from mainstream language training. We identify five recurring hallucination categories, all traceable to specific training data biases. These results suggest that current LLMs do not generate reliable TLA+ specifications without expert oversight. We release the evaluation framework, code, and dataset to support reproducibility and future research.