# model evaluation insights Rangoon Insights reporting and guides tagged model evaluation. Canonical: https://rangoon.ai/insights/tags/model-evaluation/ ## Topics - [Agent systems](https://rangoon.ai/insights/topics/agents/): Agent runtimes, handoffs, tools, and the boundaries that make them reviewable. - [Governance](https://rangoon.ai/insights/topics/governance/): Policy, identity, authorization, approvals, and evidence for consequential work. - [Infrastructure](https://rangoon.ai/insights/topics/infrastructure/): Model serving, sandboxes, tool services, and deployment boundaries. - [Models](https://rangoon.ai/insights/topics/models/): Inference choices, local serving, compatibility, and operational tradeoffs. ## Stories - [Lam introduced Fabtex Yield Optimizer for high-volume fab process variability work](https://rangoon.ai/insights/archive-lam-fabtex-yield-optimizer-october-2025/) Lam introduced Fabtex Yield Optimizer on October 7, 2025 as software using virtual silicon, inline fab data and optimization methods for yield work in high-volume manufacturing. - [OpenAI introduced GDPval for one-shot professional knowledge-work deliverables](https://rangoon.ai/insights/archive-openai-gdpval-knowledge-work-evaluation-september-2025/) OpenAI introduced GDPval on September 25, 2025, evaluating model deliverables across 44 occupations with expert comparison and a public gold subset. - [ECMWF put its 51-member AIFS ensemble into operations, with a different resolution trade-off](https://rangoon.ai/insights/archive-ecmwf-aifs-ens-operational-july-2025/) ECMWF made AIFS ENS operational on July 1, 2025, adding a 51-member machine-learning ensemble beside its IFS ensemble forecast service. - [MiniMax-M1 paired a million-token context claim with two stated thinking budgets and vendor benchmarks](https://rangoon.ai/insights/archive-minimax-m1-open-hybrid-attention-june-2025/) MiniMax introduced open-weight M1 on June 16, 2025, with a stated 1M-token context and separate 40K and 80K thinking-budget variants. - [OpenAI’s April Preparedness Framework update made evaluations, safeguards and disclosure separate controls](https://rangoon.ai/insights/archive-openai-preparedness-framework-update-april-2025/) OpenAI updated its Preparedness Framework on April 15, 2025, revising its severe-harm risk evaluation, safeguards, governance and disclosure approach for frontier capabilities. - [OpenAI released PaperBench to evaluate agents replicating AI research](https://rangoon.ai/insights/archive-openai-paperbench-ai-research-replication-april-2025/) OpenAI released PaperBench on April 2, 2025, using ICML papers, hierarchical rubrics and an open implementation to evaluate research replication. - [METR proposed task-completion time horizons for measuring long software work](https://rangoon.ai/insights/archive-metr-task-completion-time-horizons-march-2025/) METR published its time-horizon study on March 19, 2025, relating agent success to expert human task duration across multi-step tasks. - [NVIDIA released CorrDiff for regional weather downscaling as its peer-reviewed limits remained explicit](https://rangoon.ai/insights/archive-nvidia-corrdiff-weather-downscaling-february-2025/) NVIDIA announced CorrDiff on February 24, 2025 for high-resolution weather downscaling; the associated paper describes a Taiwan-focused 25 km to 2 km study. - [Epoch AI introduced FrontierMath with unpublished expert mathematics problems](https://rangoon.ai/insights/archive-epoch-frontiermath-expert-mathematics-benchmark-november-2024/) Epoch AI introduced FrontierMath on November 8, 2024, pairing original expert mathematics problems with automated answer verification. - [Aeneas made ancient-text parallels an inspectable AI research aid](https://rangoon.ai/insights/archive-aeneas-ancient-inscriptions-2025/) Google DeepMind introduced Aeneas on July 23, 2025, reporting a model for contextualizing ancient Latin inscriptions through textual and historical parallels. - [Chai-2 paired de novo antibody designs with a bounded wet-lab report](https://rangoon.ai/insights/archive-chai2-zero-shot-antibody-2025/) Chai Discovery posted its Chai-2 antibody-design study in July 2025, reporting zero-shot designs and a specified set of wet-lab tests. - [Boltz-2 put binding-affinity prediction beside structure prediction](https://rangoon.ai/insights/archive-boltz2-binding-affinity-2025/) The Boltz team introduced Boltz-2 on June 6, 2025, reporting an open model for joint biomolecular structure and binding-affinity prediction. - [Aurora made Earth-system benchmarks a boundary for forecast claims](https://rangoon.ai/insights/archive-microsoft-aurora-earth-system-2025/) Microsoft Research published Aurora in May 2025 as an Earth-system foundation model evaluated across weather, cyclone, air-quality and wave-forecasting tasks. - [RFdiffusion2 made enzyme-backbone generation a testable design hypothesis](https://rangoon.ai/insights/archive-rfdiffusion2-enzyme-design-2025/) The Institute for Protein Design introduced RFdiffusion2 on April 10, 2025, reporting a model for generating enzyme backbones from minimally specified reaction chemistry. - [DeepSeek-V3-0324 separated a post-training update from a new base model](https://rangoon.ai/insights/archive-deepseek-v3-0324-update-2025/) DeepSeek recorded V3-0324 on March 25, 2025 and said it retained the prior base model while updating post-training. Checkpoint identity and regressions still need testing. - [BioEmu-1 made protein-conformation emulation a sampling question](https://rangoon.ai/insights/archive-bioemu-protein-ensembles-2025/) Microsoft Research introduced BioEmu-1 on February 20, 2025, reporting a generative model intended to emulate protein equilibrium ensembles for research. - [Evo 2 broadened genome modeling while keeping sequence design a research claim](https://rangoon.ai/insights/archive-evo2-genome-modeling-2025/) Arc Institute introduced Evo 2 on February 19, 2025, releasing a genome-scale biological model, code, weights and training-data materials for research use. - [Google’s AI co-scientist made hypothesis ranking an auditable research step](https://rangoon.ai/insights/archive-google-ai-co-scientist-2025/) Google Research introduced an AI co-scientist on February 19, 2025, describing a Gemini-based system for generating and ranking research hypotheses and plans. - [Muse made action-conditioned world modeling a closed-loop evaluation problem](https://rangoon.ai/insights/archive-muse-wham-gameplay-world-model-2025/) Microsoft Research introduced Muse on February 19, 2025, publishing a World and Human Action Model for gameplay ideation with a bounded game-data setting. - [MatterGen separated conditional crystal proposals from material confirmation](https://rangoon.ai/insights/archive-mattergen-inorganic-materials-2025/) Microsoft Research introduced MatterGen on January 16, 2025, reporting a generative model for inorganic material structures conditioned on target properties. - [GenCast made ensemble forecast calibration a central AI-weather question](https://rangoon.ai/insights/archive-gencast-ensemble-weather-2024/) Google DeepMind introduced GenCast on December 4, 2024, reporting a diffusion-based ensemble weather model evaluated against a named operational forecast baseline. - [AlphaQubit treated quantum error decoding as an inference problem](https://rangoon.ai/insights/archive-alphaqubit-quantum-error-decoder/) Google DeepMind introduced AlphaQubit on November 20, 2024, reporting a transformer-based decoder for identifying errors in surface-code quantum experiments. ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)