Skip to content

AI Recommendation Standings / archive

AI agent frameworks: August 2026

The final August 2026 standings for “What framework should I use to build AI agents?”. These numbers are fixed and can be cited by permalink. See the current standings.

View as Markdown

AI share of voice · August 2026

585 sampled answers · 10 models

  1. 1LangChainnew29.8%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · Gem · DeepSeek V4 Flash 0731 · GLM 5.2
  2. 2LlamaIndexnew27.8%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · DeepSeek V4 Flash 0731 · GLM 5.2
  3. 3CrewAInew18.9%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · DeepSeek V4 Flash 0731 · GLM 5.2
  4. 4AutoGennew18.2%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · DeepSeek V4 Flash 0731 · GLM 5.2
  5. 5LangGraphnew17.4%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · DeepSeek V4 Flash 0731 · GLM 5.2· quiet
  6. 6Pydantic AInew12.3%
    Claude Haiku 4.5 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT · DeepSeek V4 Flash 0731 · GLM 5.2

Share of voice weights each compared model equally: it is the average, across the models, of how often each one named the tool in its own answers, so a model probed more than another can't skew it. Model chips list which AI models named it this month; the pulse is the sentiment of that month's developer mentions across Hacker News, Reddit, GitHub, Stack Overflow, Bluesky and more.

Quotable · copy freely

Averaged equally across 10 AI models, LangChain is named in 29.8% of answers about ai agent frameworks (7 of 10 name it).
Source: BackTalk AI Recommendation Standings, August 2026 · backtalk.sh/ai-recommends/ai-agent-frameworks/2026-08
The 3 prompts behind these standings
  • What framework should I use to build AI agents? (recommendation)
  • What are the best LangChain alternatives? (alternatives)
  • Which framework is best for building an LLM app? (recommendation)

Each prompt is sampled several times per model per probe because answers vary run to run; 585 answers across 8 probe runs fed this month.