# Evaluation insights Rangoon Insights reporting and guides tagged Evaluation. Canonical: https://rangoon.ai/insights/tags/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 - [Anthropic released Petri for broad alignment audits, while its generated transcripts still needed human interpretation](https://rangoon.ai/insights/archive-anthropic-petri-open-source-auditing-october-2025/) Anthropic released Petri on October 6, 2025 as an open-source auditing tool that runs simulated multi-turn interactions and scores the resulting model behavior for review. - [Microsoft put PyRIT-backed AI red teaming into Foundry preview, making scan evidence a separate release artifact](https://rangoon.ai/insights/archive-microsoft-ai-red-teaming-agent-preview-april-2025/) Microsoft announced its AI Red Teaming Agent public preview on April 4, 2025, bringing PyRIT-based adversarial testing to Foundry evaluation, reporting and logging surfaces. - [GPT-5.2 separated work modes without making their results interchangeable](https://rangoon.ai/insights/archive-openai-gpt-5-2-december-2025/) OpenAI introduced GPT-5.2 on December 11, 2025, describing Instant, Thinking, and Pro variants for different kinds of work. - [SWE-Bench Pro brought private and long-horizon code tasks into agent evaluation](https://rangoon.ai/insights/archive-swe-bench-pro-september-2025/) Scale AI introduced SWE-Bench Pro on September 19, 2025, a 1,865-problem coding-agent benchmark spanning public, held-out, and commercial repositories. - [Mistral Medium 3 framed model choice around cost and deployment claims](https://rangoon.ai/insights/archive-mistral-medium-3-2025/) Mistral introduced Medium 3 on May 7, 2025. Its benchmark, price, and deployment claims require a workload-specific quality and cost basis. - [ARC-AGI-2 added calibrated task sets and a cost axis to its reasoning challenge](https://rangoon.ai/insights/archive-arc-agi-2-march-2025/) ARC Prize launched ARC-AGI-2 on March 24, 2025, with grid-based reasoning tasks, 120-task evaluation sets, pass@2 scoring, and cost reporting. - [Humanity’s Last Exam released 3,000 expert-crafted questions for a hard academic test](https://rangoon.ai/insights/archive-humanitys-last-exam-january-2025/) Scale AI and the Center for AI Safety released Humanity’s Last Exam on January 23, 2025: 3,000 multimodal, closed-ended academic questions. - [The DeepSeek-R1 paper made the training path part of the story](https://rangoon.ai/insights/archive-deepseek-r1-paper-submission-2025/) DeepSeek submitted its DeepSeek-R1 paper to arXiv on January 22, 2025. Its account of reinforcement learning and distillation called for claim-by-claim model evaluation. - [FACTS Grounding’s January paper defined a two-stage test for document-faithful answers](https://rangoon.ai/insights/archive-deepmind-facts-grounding-january-2025/) Google DeepMind’s January 6, 2025 FACTS Grounding paper described a 1,719-example benchmark for long-form answers tied to supplied documents. - [DeepSeek-V3’s December 2024 API update made a checkpoint date operational](https://rangoon.ai/insights/archive-deepseek-v3-api-upgrade-2024/) DeepSeek documented a DeepSeek-V3 upgrade for deepseek-chat on December 26, 2024. A stable endpoint name does not preserve checkpoint identity, prompt behavior, or evaluation. - [Phi-4’s December 2024 release put data quality in the model record](https://rangoon.ai/insights/archive-microsoft-phi-4-data-quality-2024/) Microsoft introduced Phi-4 on December 12, 2024 as a 14-billion-parameter small language model. Its training account made data and curriculum review part of evaluation. - [Amazon Nova’s 2024 launch made model-family choice a workload decision](https://rangoon.ai/insights/archive-amazon-nova-family-launch-2024/) Amazon introduced the Nova family on December 3, 2024. Its text, multimodal, image, and video models made an input-output map more useful than one family label. - [Claude Frontier Academy: assess deployment skills beyond the credential](https://rangoon.ai/insights/claude-frontier-academy-deployment-evidence/) Anthropic’s engineer-training commitment raises a practical question for development firms: what evidence shows that an AI workflow is ready for a client handoff? ## 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)