--- title: "State of LLMs on the Application Layer" description: "We have analyzed model adoption of 20.000+ organizations building LLM applications and agents in the last 12 months. Here is what we found." ogImage: /images/blog/2025-10-13-state-of-llms-september-2025/state-of-llms-september-2025.png tag: update date: 2025/10/13 author: "felixkrauth" --- ## About Langfuse Langfuse is the [leading open source LLM observability platform](/blog/2024-11-most-used-oss-llmops) powering safe and scalable LLM applications for 10s of thousands of organizations. The data on Langfuse Cloud tells a story of a rapidly growing market and allows us to get a unique view on the state of the LLM application layer. We power global enterprises like [Khan Academy](/users/khan-academy), [SumUp](/users/sumup) and [Merck](/users/merckgroup) as well as AI-native startups like [Magic Patterns](/users/magic-patterns-ai-design-tools), [Circleback](https://circleback.ai/) and [Juicebox](https://juicebox.ai/) to build safe and scalable LLM applications and agents. ## About this Report The core of what we do is helping developers to ship LLM applications and agents faster and more reliably. Our dataset gives a **unique view on how models are adopted on the application layer** instead of consumers that use those models via apps like Claude, Gemini or ChatGPT. - The data comes from **over 20.000 organizations** on Langfuse Cloud - **Billions of LLM traces and observations** ([see data model](/docs/observability/data-model)) on Langfuse Cloud (no self-hosted data included) - **59 unique models tracked across major providers**, longtail of models is not included and grouped into "Other" - **Time Period:** 12 months (Oct 2024 - Sep 2025) ## Status Quo in September 2025 Here is where the market stands in September 2025: OpenAI is still the dominant player (55.3%), followed by Google (13.1%) and Anthropic (7.3%). The longtail of models is not included and grouped into "Other" (24.3%). ![Status Quo in September 2025](/images/blog/2025-10-13-state-of-llms-september-2025/Marketshare-sep.png) ## Change over the last 12 months The past 12 months show clear changes in the market. ![Status Quo in September 2025](/images/blog/2025-10-13-state-of-llms-september-2025/Marketshare-12mo.png) | Provider | Sep 2025 | Oct 2024 | Change | | --------- | -------- | -------- | --------- | | OpenAI | 55.3% | 82.7% | -27.4 pts | | Other | 24.3% | 10.0% | +14.3 pts | | Google | 13.1% | 0.5% | +12.6 pts | | Anthropic | 7.3% | 6.8% | +0.5 pts | ## Key Insights - **OpenAI is yielding market share** (down -27.4 points) while Google, Anthropic and "Other" model providers grow. - **Google's Rise:** Google went from 0.5% → 13.1% (26x) share in just 12 months. - **Anthropic is relatively stable** at 7.3% - **"Other" models rising** strongly signals that the LLM market is fragmenting beyond the "Big Three," with organizations adopting specialized models, open-source alternatives, and regional providers. ## Market Share by Model Let's take a look at market shares on model level. ![Market Share by Model](/images/blog/2025-10-13-state-of-llms-september-2025/top-models-over-time.png) **What we see** - The market is operating on a **~6 month effective model lifecycle** – Several October 2024 top-20 models effectively disappeared - **Application builders are heavily relying on small efficient models** (OpenAI nano or mini: Gemini Flash; Anthropic’s Opus model not relevant) - The **story isn't "OpenAI vs everyone else"—it's fragmentation vs. consolidation**. While the Big Three (OpenAI, Google, Anthropic) consolidate around 75% of tracked share, the "Other" 25% represents hundreds of specialized models serving specific use cases, geographies, or price points. **Key Takeaways:** - **Organizations aren't building on stable platforms**; they're building on a constantly shifting foundation. This creates both opportunity (rapid innovation) and risk (technical debt accumulates fast). - **Staying flexible and being ready to swap out models will pay off** when building applications and underlying infrastructure. ## Top 15 Models - September 2025 1. GPT-4o mini (OpenAI): 14.2% 2. GPT-4o (OpenAI): 10.9% 3. GPT-4.1 (OpenAI): 9.8% 4. gemini-2.5-flash (Google): 9.4% 5. GPT-4.1 mini (OpenAI): 8.3% 6. gemini-2.0-flash (Google): 3.7% 7. Claude Sonnet 4 (Anthropic): 3.6% 8. GPT-5 nano (OpenAI): 3.4% 9. GPT-5 mini (OpenAI): 3.2% 10. GPT-5 (OpenAI): 1.7% 11. Claude Sonnet 3.7 (Anthropic): 1.5% 12. GPT-4.1 nano (OpenAI): 1.0% 13. o3 (OpenAI): 0.9% 14. o4-mini (OpenAI): 0.6% 15. Claude Sonnet 3.5 (Anthropic): 0.6% ## Top 15 Models - 12 Month Average 1. GPT-4o mini (OpenAI): 24.2% 2. GPT-4o (OpenAI): 16.4% 3. GPT-4.1 mini (OpenAI): 6.6% 4. GPT-4.1 (OpenAI): 6.3% 5. gemini-2.0-flash (Google): 5.3% 6. gemini-2.5-flash (Google): 4.4% 7. Claude Sonnet 4 (Anthropic): 2.0% 8. GPT-4.1 nano (OpenAI): 1.8% 9. Claude Sonnet 3.7 (Anthropic): 1.8% 10. Claude Sonnet 3.5 (Anthropic): 1.2% 11. GPT-5 nano (OpenAI): 1.1% 12. GPT-5 mini (OpenAI): 0.8% 13. Claude Haiku 3 (Anthropic): 0.8% 14. text-embedding-3-large (OpenAI): 0.7% 15. text-embedding-ada-002 (OpenAI): 0.7% --- **Data source:** Langfuse LLM observability platform\ **Analysis period:** October 2024 - September 2025 _Want to reach out to talk about this story? Write us at press@langfuse.com._ --- ## Agent Instructions This page is part of the [Langfuse](https://langfuse.com) documentation, published as plain Markdown for AI agents. Every page is available as Markdown by appending `.md` to its URL, or by sending an `Accept: text/markdown` header. This page: `https://langfuse.com/blog/2025-10-13-state-of-llms-september-2025.md`. ### Querying these docs If the answer is not on this page, query the documentation instead of guessing: - **Semantic search** across all Langfuse docs, returning an answer with the relevant pages and excerpts. Ask a specific, self-contained question: ```bash curl -sG "https://langfuse.com/api/search-docs" --data-urlencode "query=How do I trace a LangGraph agent?" ``` - **Index of every page**: , with per-section indexes [llms-docs.txt](https://langfuse.com/llms-docs.txt), [llms-integrations.txt](https://langfuse.com/llms-integrations.txt), and [llms-self-hosting.txt](https://langfuse.com/llms-self-hosting.txt). ### Before writing Langfuse code - **Install the [Langfuse Agent Skill](https://langfuse.com/docs/api-and-data-platform/features/agent-skill).** It encodes Langfuse's own best practices for instrumentation, prompt management, and evaluation, and materially improves results. - **Read [What does a good trace look like?](https://langfuse.com/docs/observability/best-practices.md)** before instrumenting an application. - **Verify endpoints, parameters, and response fields** against the [API reference](https://api.reference.langfuse.com) instead of inferring them from code examples. - **Use the [Langfuse CLI](https://langfuse.com/docs/api-and-data-platform/features/cli)** (`npx langfuse-cli api `) to read or write traces, prompts, datasets, and scores from the terminal. Found an error in these docs? Please open an issue at .