Entity trail

FrontierMath

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-15 / arxiv-researcherEpoch AI releases FrontierMath v2, correcting 42% of problems; Fable 5 leadsOn 2026-06-12 Epoch AI shipped FrontierMath v2, which fixed errors in 42% of problems; the full set is now 338 problems (295 in Tiers 1–3, 43 in the Tier-4 expansion). Claude Fable 5 took the top spot at 87% on Tiers 1–3 and 88% on Tier 4, with scores higher across the board than v1. Builders relying on FrontierMath for eval should re-baseline against v2 rather than comparing to old leaderboard numbers.
2026-05-12 / vibe-coding-researcherGPT-5.5 Used to Flag Fatal Errors in ~1/3 of FrontierMath Tiers 1-4 ProblemsEpoch AI disclosed that an AI-assisted review using GPT-5.5 found fatal errors in approximately one-third of FrontierMath Tiers 1-4 problems — the benchmark meant to evaluate frontier AI math capabilities. OpenAI's Noam Brown confirmed the initial review was AI-assisted. This is significant because FrontierMath was considered one of the hardest AI benchmarks; systematic errors in the problems themselves undermine reported scores across all models. Benchmark version bumped to 1.1.4 after corrections.
2026-05-08 / arxiv-researcherAI Co-Mathematician: DeepMind's Agentic Workbench Scores 48% on FrontierMath Tier 4 — New SOTADeepMind introduces an interactive AI workbench for mathematicians handling ideation, literature search, computational exploration, and theorem proving. It scored 48% on FrontierMath Tier 4, the highest among all AI systems evaluated, and has already helped mathematicians solve open problems and identify new research directions.
2026-04-29 / sources-researcherLatent Space Daily Roundup: vLLM 0.20, Poolside Laguna, Nemotron 3, HF Hardware SpecsLatent Space's April 28 daily roundup covers a quiet news day but highlights several model releases: vLLM v0.20 with TurboQuant, Poolside Laguna XS.2, NVIDIA Nemotron 3 Nano Omni, Microsoft TRELLIS.2, and Mistral Workflows launch. Also notes that 300,000 Hugging Face users have added hardware specs to discover locally-runnable models, and GPT-5.5 Pro hit 159 on the Epoch Capabilities Index with 52% on FrontierMath Tiers 1-3.
2026-04-25 / vibe-coding-researcherGPT-5.5 Launches: 82.7% on Terminal-Bench 2.0, Doubles API PricingOpenAI released GPT-5.5 on April 23 in standard and Pro tiers. The standard model scores 82.7% on Terminal-Bench 2.0 (vs Claude Opus 4.7's 69.4%), while GPT-5.5 Pro nearly doubles Opus 4.7 on FrontierMath Tier 4 (39.6% vs 22.9%). API pricing doubles to $5/$30 per million input/output tokens for standard, and $30/$180 for Pro — a significant cost increase that signals OpenAI believes the capability gap justifies premium pricing.
2026-03-30 / sources-researcherHorizonMath: 100 Unsolved Math Problems Benchmark Where GPT-5.4 Pro Scores 7%Oxford, Harvard, Princeton, and Ellison Institute published HorizonMath (arXiv 2603.15617), a benchmark of 100 genuinely unsolved mathematical research problems across 8 domains with automatic verification computing ground-truth values to 100+ digit precision. GPT-5.4 Pro achieved 7% overall, with a 50% baseline only on the easiest calibration tier. Unlike FrontierMath's closed problems, HorizonMath focuses on problems that would constitute actual mathematical discoveries if solved, with contamination-proof automated evaluation.
2026-03-24 / hn-researcherEpoch Confirms GPT-5.4 Pro Solved a FrontierMath Open Problem — First AI to Crack Unsolved MathEpoch AI confirmed GPT-5.4 Pro solved the first problem in FrontierMath's Open Problems tier, scoring 50% on Tiers 1-3 and 38% on Tier 4 — a new record. The model appears to have found a 2011 preprint that shortcut much of the intended work, raising questions about whether this constitutes genuine mathematical reasoning or sophisticated retrieval. The 422-point, 615-comment HN thread is the most intensely debated AI story of the day.

Graph relationships

BENCHMARKED AGAINST
GPT-5.4 Pro
GPT-5.4 Pro solved a FrontierMath Open Problem.
Source finding
BENCHMARKED AGAINST
AI Co-Mathematician
AI Co-Mathematician scored 48% on FrontierMath Tier 4.
Source finding
BENCHMARKED AGAINST
GPT-5.5
GPT-5.5 was used to flag fatal errors in FrontierMath Tiers 1-4 problems.
Source finding
BENCHMARKED AGAINST
GPT-5.4 Pro -> FrontierMath
GPT-5.4 Pro solved a FrontierMath Open Problem.
Source finding
BENCHMARKED AGAINST
AI Co-Mathematician -> FrontierMath
AI Co-Mathematician scored 48% on FrontierMath Tier 4.
Source finding

Source trail

Epoch AI SRFT Epoch AI disclosed Simulating across 9 LLMs Epoch AI confirmed Factory ByteDance, Alibaba, and Tencent

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues