# open weights insights

Rangoon Insights reporting and guides tagged open weights.

Canonical: https://rangoon.ai/insights/tags/open-weights/

## 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
- [GLM-4.7 made preserved thinking a named control for multi-turn coding agents](https://rangoon.ai/insights/archive-zai-glm-4-7-december-2025/)
  Z.ai released GLM-4.7 on December 22, 2025, pairing open model weights with interleaved, preserved, and turn-level thinking controls for agent workflows.
- [Tencent released Hunyuan-A13B under a territory-limited community license, separating open weights from universal permission](https://rangoon.ai/insights/archive-tencent-hunyuan-a13b-open-hybrid-reasoning-june-2025/)
  Tencent’s June 27, 2025 Hunyuan-A13B release paired 80B-total and 13B-active MoE claims with a territory-limited community license.
- [Kimi K2-0905 made context, tool parsing, and serving configuration one release surface](https://rangoon.ai/insights/archive-moonshot-kimi-k2-0905-september-2025/)
  Moonshot AI announced Kimi K2-0905 on September 5, 2025, citing a 256K context window, enhanced frontend and tool-calling work, and broader agent-scaffold integration.
- [DeepSeek-V3.1 joined thinking modes, long context, and tool calls in one release](https://rangoon.ai/insights/archive-deepseek-v3-1-august-2025/)
  DeepSeek released DeepSeek-V3.1 on August 21, 2025, describing Think and Non-Think modes, 128K context, updated tokenizer materials, and beta strict function calling.
- [GLM-4.5 put hybrid reasoning modes and open weights in one release record](https://rangoon.ai/insights/archive-zai-glm-4-5-july-2025/)
  Z.AI introduced GLM-4.5 and GLM-4.5-Air on July 28, 2025, describing hybrid reasoning modes, agentic capabilities, and open-weight distribution.
- [Llama 4 showed why open weights still need deployment boundaries](https://rangoon.ai/insights/archive-llama-4-open-weights-boundary/)
  Meta released Llama 4 Scout and Maverick in April 2025. Their multimodal, mixture-of-experts design made model cards, licensing and workload conditions part of adoption evidence.
- [Mistral Small 3.1 made multimodal acceptance a modality-by-modality test](https://rangoon.ai/insights/archive-mistral-small-3-1-2025/)
  Mistral released Small 3.1 on March 17, 2025 as an updated small model. A vision-capable release needs image, text, retrieval, and structured-output tests kept separate.
- [Mistral Small 3 made latency a claim to measure in the target stack](https://rangoon.ai/insights/archive-mistral-small-3-latency-2025/)
  Mistral introduced Mistral Small 3 on January 30, 2025 as a 24B Apache 2.0 model optimized for low latency. A rate claim needs a reproducible serving profile.
- [Llama 3.3’s 70B release made license and benchmark context part of adoption](https://rangoon.ai/insights/archive-llama-3-3-70b-2024/)
  Meta released Llama 3.3 70B Instruct on December 6, 2024. Open weights, a model card, and vendor evaluations each answer different adoption questions.
- [Qwen2.5-Coder’s family release made coding-model selection a size-and-harness decision](https://rangoon.ai/insights/archive-qwen2-5-coder-open-code-models-2024/)
  Qwen released the Qwen2.5-Coder series on November 12, 2024, expanding open code models across six sizes and describing assistant and artifact scenarios.
- [Aya Expanse made multilingual evaluation a release requirement, not a translation claim](https://rangoon.ai/insights/archive-cohere-aya-expanse-multilingual-2024/)
  Cohere For AI launched Aya Expanse on October 24, 2024, releasing 8B and 32B multilingual models aimed at narrowing language performance gaps.


## 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)