# Folkbench

> Folkbench is an independent directory for comparing AI relay services using published ranking, availability, pricing, privacy, and operational evidence.

## Primary pages
- Home: https://folkbench.com/
- Rankings: https://folkbench.com/rankings
- Model catalog: https://folkbench.com/models
- Blog: https://folkbench.com/blog
- Pelican gallery: https://folkbench.com/pelican
- Trust and security: https://folkbench.com/trust
- Vendor participation: https://folkbench.com/vendor-join

## Machine-readable discovery
- Full overview: https://folkbench.com/llms-full.txt
- Sitemap: https://folkbench.com/sitemap.xml
- Robots policy: https://folkbench.com/robots.txt

Ranking and station pages contain publication timestamps and evidence limitations. Do not treat a sampled result as a guarantee of future service quality.

## Published articles
- [How to evaluate an AI relay for software engineering](https://folkbench.com/en/blog/ai-api-relay-software-engineering-evaluation-guide) (en)
- [Stop trusting a 99% uptime claim: the AI API relay failures we hit, and how we test stability](https://folkbench.com/en/blog/ai-api-relay-stability-guide) (en)
- [From translationese to a felt native voice: three years of Chinese writing with AI, and with API relays](https://folkbench.com/en/blog/ai-chinese-writing-models-api-relay-user-submission) (en)
- [API relay reliability checks](https://folkbench.com/en/blog/api-relay-reliability-checks) (en)
- [Whether an AI can handle workplace talk is not about a polished line. It is about turning a snide remark into facts, a boundary, and a next step](https://folkbench.com/en/blog/boss-yinyang-workplace-reply) (en)
- [How Strong Is Haiku 5.5, Really? On the Official Scorecard, First Place Is Not the Part Worth Reading](https://folkbench.com/en/blog/claude-haiku-5-5-official-scorecard) (en)
- [Claude Opus 5.5 specs, list prices, and published benchmarks](https://folkbench.com/en/blog/claude-opus-5-5-review-pricing) (en)
- [Claude Sonnet 5.5 specs, list prices, and published benchmarks](https://folkbench.com/en/blog/claude-sonnet-5-5-review-pricing) (en)
- [Did the Model Really Get Dumber? From One Failed Screenshot to a Credible Comparison](https://folkbench.com/en/blog/did-the-model-get-dumber) (en)
- [GPT-6.1 Sol specs, long-context prices, and published comparisons](https://folkbench.com/en/blog/gpt-6-1-sol-review-pricing) (en)
- [Grok 4.7 specs, long-context prices, and published benchmarks](https://folkbench.com/en/blog/grok-4-7-review-pricing) (en)
- [How Do You Score a Pelican on a Bicycle? From a Viral Test to a Credible Evaluation System](https://folkbench.com/en/blog/how-to-score-a-cycling-pelican) (en)
- [Can it cook, or only write the recipe? Twelve home-kitchen scenes show whether a model's steps can be carried out](https://folkbench.com/en/blog/jianbing-guozi-executable-steps) (en)
- [Kimi K3.1 Pro in depth: long coding tasks, 3D work and real operating costs](https://folkbench.com/en/blog/kimi-k3-1-pro-deep-dive) (en)
- [The gray-market ghost in model relays: founders burned by a 10% API](https://folkbench.com/en/blog/llm-api-relay-gray-industry-user-submission) (en)
- [When “don’t spoil the mood” beats telling the truth: praising mom’s cooking exposes a blind spot in AI EQ tests](https://folkbench.com/en/blog/mom-dish-praise-honesty) (en)
- [Cómo evaluar un intermediario de IA para ingeniería de software](https://folkbench.com/es/blog/ai-api-relay-software-engineering-evaluation-guide) (es)
- [Deja de fiarte de un 99% de disponibilidad: los fallos de relay de API de IA que nos encontramos, y cómo probamos la estabilidad](https://folkbench.com/es/blog/ai-api-relay-stability-guide) (es)
- [Del chino traducido a una voz que se siente propia: tres años escribiendo en chino con IA y con relays de API](https://folkbench.com/es/blog/ai-chinese-writing-models-api-relay-user-submission) (es)
- [Comprobaciones de fiabilidad de un relay de API](https://folkbench.com/es/blog/api-relay-reliability-checks) (es)
- [Si una IA sabe comunicarse en el trabajo no se ve en la frase bonita, sino en si convierte un comentario cargado en hechos, un límite y el siguiente paso](https://folkbench.com/es/blog/boss-yinyang-workplace-reply) (es)
- [¿Qué tan fuerte es de verdad Haiku 5.5? En el boletín oficial, lo que más merece verse no es el primer puesto](https://folkbench.com/es/blog/claude-haiku-5-5-official-scorecard) (es)
- [Claude Opus 5.5: especificaciones, precios de lista y benchmarks publicados](https://folkbench.com/es/blog/claude-opus-5-5-review-pricing) (es)
- [Especificaciones, precios de lista y benchmarks publicados de Claude Sonnet 5.5](https://folkbench.com/es/blog/claude-sonnet-5-5-review-pricing) (es)
- [¿El modelo de verdad se volvió más tonto? De una captura fallida a una comparación creíble](https://folkbench.com/es/blog/did-the-model-get-dumber) (es)
- [GPT-6.1 Sol: especificaciones, precios de contexto largo y comparaciones publicadas](https://folkbench.com/es/blog/gpt-6-1-sol-review-pricing) (es)
- [Especificaciones de Grok 4.7, precios de contexto largo y benchmarks publicados](https://folkbench.com/es/blog/grok-4-7-review-pricing) (es)
- [¿Cómo se puntúa un pelícano en bicicleta? De una prueba viral a un sistema de evaluación creíble](https://folkbench.com/es/blog/how-to-score-a-cycling-pelican) (es)
- [¿Sabe cocinar, o solo escribir la receta? Doce escenas de cocina casera muestran si los pasos se pueden ejecutar](https://folkbench.com/es/blog/jianbing-guozi-executable-steps) (es)
- [Kimi K3.1 Pro a fondo: programación prolongada, creación 3D y costes reales](https://folkbench.com/es/blog/kimi-k3-1-pro-deep-dive) (es)
- [El fantasma del mercado gris en los relays de modelos: fundadores quemados por una API al 10%](https://folkbench.com/es/blog/llm-api-relay-gray-industry-user-submission) (es)
- [Cuando “no aguar la fiesta” pesa más que decir la verdad: elogiar la comida de mamá muestra un punto ciego](https://folkbench.com/es/blog/mom-dish-praise-honesty) (es)
- [写代码场景下，AI 中转站的软件工程评测指南](https://folkbench.com/blog/ai-api-relay-software-engineering-evaluation-guide) (zh-CN)
- [别再被 99% 可用率骗了：我们踩过的 AI API 中转站大坑与稳定性测试指南](https://folkbench.com/blog/ai-api-relay-stability-guide) (zh-CN)
- [从“翻译腔”到“精神母语”：三年 AI 中文写作与 API 中转站体验](https://folkbench.com/blog/ai-chinese-writing-models-api-relay-user-submission) (zh-CN)
- [API 中转站可靠性检查](https://folkbench.com/blog/api-relay-reliability-checks) (zh-CN)
- [AI 真正会不会职场沟通，不看会不会说漂亮话，而看能不能把“阴阳话”翻译成事实、边界和下一步](https://folkbench.com/blog/boss-yinyang-workplace-reply) (zh-CN)
- [Haiku 5.5 到底有多强？官方成绩单里，最值得看的不是第一名](https://folkbench.com/blog/claude-haiku-5-5-official-scorecard) (zh-CN)
- [Claude Opus 5.5 的规格、标价与发布页基准](https://folkbench.com/blog/claude-opus-5-5-review-pricing) (zh-CN)
- [Claude Sonnet 5.5 的规格、标价与发布页基准](https://folkbench.com/blog/claude-sonnet-5-5-review-pricing) (zh-CN)
- [模型真的“降智”了吗？从一张翻车截图到一场可信的对照实验](https://folkbench.com/blog/did-the-model-get-dumber) (zh-CN)
- [GPT-6.1 Sol 的规格、长上下文标价与发布页比较](https://folkbench.com/blog/gpt-6-1-sol-review-pricing) (zh-CN)
- [Grok 4.7 的规格、长上下文标价与发布页基准](https://folkbench.com/blog/grok-4-7-review-pricing) (zh-CN)
- [如何给一只骑车的鹈鹕打分？从网红测试到可信评测系统](https://folkbench.com/blog/how-to-score-a-cycling-pelican) (zh-CN)
- [会做菜，还是会写菜谱？12 个家庭厨房场景暴露模型的可执行性](https://folkbench.com/blog/jianbing-guozi-executable-steps) (zh-CN)
- [Kimi K3.1 Pro 深度解析：长程编程、3D 能力与真实使用成本](https://folkbench.com/blog/kimi-k3-1-pro-deep-dive) (zh-CN)
- [大模型中转站的“灰产幽灵”：那些被 1 折 API 毁掉的创业者们](https://folkbench.com/blog/llm-api-relay-gray-industry-user-submission) (zh-CN)
- [当“不扫兴”压过“说实话”：夸妈妈做菜测试暴露 AI 情商评测的盲区](https://folkbench.com/blog/mom-dish-praise-honesty) (zh-CN)