Literature search, paper writing, lab notes, academic typesetting
Research is the field where AI help is most asymmetric: retrieval, review, writing, typesetting and figures - the expression layer - have been substantially accelerated, while posing a good question, designing a falsifiable experiment and judging whether a result is credible - the judgement layer - have barely been touched. That asymmetry matters because it fixes where the tools belong: use AI to read more and write faster, not to decide what is worth studying. The link to AGI is that research is one of the few domains with expensive but well-defined verification, so a system that can close the hypothesis-experiment-check loop autonomously means far more than a model that can produce a paper. That is why agent-based research workbenches, rather than paper generators, are what to watch. The ladder here is ranked by whether a tool advances that loop, not by how much its output resembles a paper.
Tracked assets
3
Installable skills
3
Capability domains
13
SOTA applications
3
MathResearchTopB
AlphaEvolve: pointing a model at problems that come with an objective scoring function
Google DeepMind's coding-agent-driven evolutionary search: Gemini Flash proposes candidates in volume and Gemini Pro in quality, automatic evaluators score them, and high scorers stay in the population as next-round context, i.e. evolutionary search with an LLM as the mutation operator; it applies only where an automatic evaluator exists, hence the math and agents filing. An evolved scheduling heuristic has run in production in Google data centres (Borg) for over a year, recovering 0.7% of Google's global compute (Google's compute, not all compute on earth), the card reading and the only result validated by long-running production. An evolved matmul kernel is 23% faster at specific sizes, and about 20% of 50+ open maths problems improved, including 4x4 complex matrix multiplication in 48 multiplications against Strassen's 1969 record of 49. Verification differs from the prover line: a Lean check is mathematical correctness, "23% faster" is an empirical reading on specific hardware. No public weights, Early Access is a waitlist, and the precondition is an evaluator you write yourself. Graded B (contested): a year of production behind the Borg result, nothing reproduced.
0.7%Google global compute recovered (production-verified)Contested · 2025-05
DeepSeek-Prover-V2: making the Lean kernel, not the model, the judge of correctness
DeepSeek's formal theorem proving model: a 671B MoE base shared with DeepSeek-V3 and Lean 4 as target language, so formal alignment on a flagship base. The core move is recursive subgoal decomposition at cold start: one-shot proofs of competition statements succeed so rarely that RL has almost no positive samples, so DeepSeek-V3 splits a statement into subgoals, a 7B model proves each, and the composed proofs become training data, trading decomposability for data density; the RL reward is binary and comes from the Lean kernel. Vendor-reported: MiniF2F-test 88.9% at pass@8192, PutnamBench 49/658 (about 7.4%), own ProverBench 325 problems. Read them together: MiniF2F is near saturation at that budget while hard Putnam problems drop an order of magnitude, and ProverBench is self-built and self-run. The public repo (about 1304 stars) ships a ZIP of proofs you can run through Lean, putting verifiability a tier above closed models. Limits: pass@8192 means eight thousand-plus candidates per problem, not comparable with pass@1. Not commensurable with Kimina-Prover's 92.2%, so list both without ranking. Graded C (vendor-stated): nothing recomputed in a standard Lean 4 + mathlib environment.
Kimina-Prover: reaching 92.2% on miniF2F with whole-proof RL and no search tree
The Lean 4 theorem proving line from Moonshot AI with Numina, deliberately contrarian: no prover feedback in training or inference, no Monte Carlo tree search, value function or process reward model (arXiv:2504.11354), so search is internalised into a single whole-proof generation, and a custom Formal Reasoning Pattern writes a natural-language draft before Lean code, recasting a sparse binary reward as a language problem. The base is Qwen2.5-72B with up to 32K context. Keep two generations apart: Preview (2025-04) reached 80.7% on miniF2F-test at pass@8192, the first public result above 80%; the 72B model (2025-07-10) reached 92.2% after adding TTRL search and error repair, the highest public reading then and the number on the card, while pass@1 under the same protocol is only 63.9%. Equal-budget pass@1/32/1024 of 63.9/84.0/87.7 leads the 671B DeepSeek-Prover-V2 (61.9/82.4/86.6) at all three budgets. Discounts: that table is Kimina's own and its DeepSeek rows are their reruns, the repo declares no licence, and miniF2F holds 244 problems now past 92%, so its power to separate frontier models is decaying. No third-party board exists here, so read the numbers as direction. Graded C (vendor-stated): nothing recomputed on a pinned commit.
Academic Research Skills: A five-skill research pipeline
Editor's takeIts strongest idea is not that it can draft a paper, but that it makes research state, evidence, and stopping conditions explicit: every stage needs scholar confirmation, integrity gates inspect citations and declared provenance, and simulated review is constrained to author feedback rather than a publication decision.
v3.23.0, 50,461 stars, CC BY-NC 4.0. Five Claude Code skills connect deep research, paper writing, simulated peer review, a 10-stage orchestrator, and systematic-review screening, using a Material Passport, mandatory checkpoints, and integrity gates to keep the scholar in control.
Three Chinese-first research skills that run on both Claude Code and Codex
Editor's takeThe part worth copying is that academic integrity is written as instructions: no fabricated data, DOIs or journals, and every piece of content is classified as quotation, confirmed fact, reasonable inference or writing suggestion, so the author can see at a glance which sentences have a source. The real danger in academic writing is not weak prose but a model writing inference as though it were established fact, and that four-way split blocks exactly that. The three skills also compose: figures, then body text, then defence slides.
4,117 stars, MIT. research-writing handles drafting, revising and polishing paper sections plus rebuttals; office-academic generates and edits academic Word and PPT at the OOXML layer (literature reports, group meetings, proposals, defences); scientific-toolkit covers MATLAB/Python simulation, signal processing, statistics, machine learning and paper figures, with literature search across arXiv, PubMed, CrossRef, Semantic Scholar and OpenAlex.
An eight-phase deep research pipeline with an evidence ledger
Editor's takeThe soul is four append-only JSONL ledgers: sources, evidence, claims and run_manifest. Every factual claim must carry an inline [N] backed by evidence.jsonl, unsupported claims may not ship, and validate_report.py plus verify_citations.py check the result. The repo has no LICENSE file, which under default copyright means all rights reserved: you can read and run it locally, but get the author's permission before redistributing or using it commercially.
1,051 stars. Four intensity modes — Quick at 3 phases and 2-5 minutes, Standard at 6, Deep at 8, UltraDeep at 8+ and 20-45 minutes — running SCOPE, PLAN, RETRIEVE, TRIANGULATE, outline refinement, SYNTHESIZE, CRITIQUE, REFINE and PACKAGE, and emitting Markdown, a McKinsey-style HTML report and a PDF.
