LLM API for Astro: what 27 makers' products chose
Among 27 makers' products built with Astro that we know the LLM API of, OpenAI API leads: 17 of them (63%). Next come Gemini API (6) and Claude (4).
Which provider, and whether to go through a gateway. Going direct to a provider gets its newest models and features first and the lowest latency; a gateway gives you one key and one bill for hundreds of models, with fallbacks when a provider is down, for a small markup or a proxy you run. The code that calls the model — an AI SDK or agent framework — is its own decision.
What makers on Astro chose
| Option | Astro products | All makers | Free for | At 10,000 users |
|---|---|---|---|---|
| 17 (63%) | 75% | ~1,000 users | $10/moGPT-6 Luna | |
| 6 (22%) | 10% | ~360 users | $28/moGemini 3.1 Flash-Lite | |
| 4 (15%) | 26% | ~1,000 users | $10/moClaude Haiku 5.5 | |
| 2 (7%) | 4% | ~1,500 users | $6.75/moGPT OSS 20B | |
| 2 (7%) | 1% | ~110 users | — | |
| 2 (7%) | 7% | — | — | |
| 1 (4%) | 2% | ~1,700 users | $6.00/moMinistral 3 (3B) | |
| 1 (4%) | 2% | — | — |
Live makers' products with a source for both: that they're built with Astro, and what they use for LLM API. A product using two options counts in both; not market share. "All makers" is the option's share among every maker's product that made this decision.
Costs are the cheapest plan that fits, from each vendor's pricing page, at the usage our stack planner assumes for a product with that many users. Free for: the users its cheapest plan a business can use stays free for; plans for personal use only, like a non-commercial hobby tier, don't count.
Which to pick, by situation
- You want the strongest general models and the widest ecosystem of examples and integrations
OpenAI API 17 here
Claude 4 here
Gemini API 6 here - Token cost dominates your budget, for example high-volume batch or agent workloads
Groq 2 here
Mistral AI 1 here - You want to try or mix many models with one key and one bill, and fall back when a provider is down
OpenRouter 2 here - You want that routing and spend control but self-hosted, with keys and logs on your servers
LiteLLM 1 here
OpenAI API
Model provider17 Astro products here (63%). Free up to about 1,000 users. About $10 a month at 10,000 users (GPT-6 Luna).
llms.txtIn Lovable, Replit, RorkBest forThe broadest model lineup — text, images, speech, embeddings — and the largest ecosystem.
Trade-offFrequent model and API changes to keep up with, and no option to run its hosted models elsewhere.
“Whisper delivers state-of-the-art speech recognition with unmatched accuracy, enabling seamless voice interactions in our AI-powered websites and apps.”
- Its models are strong at code generation, reasoning and creative tasks, often chosen after testing alternatives. 4 sources
- The API is reliable and well-priced as the foundation for production features. 4 sources
- It cut usage on the $200 Pro plan and changed limits, frustrating paying subscribers. 4 sources
- Model quality shifts between releases, including a quickly replaced GPT-6 and quietly tweaked reasoning levels. 4 sources
Status page: 5+ incidents in the last 90 days (its own page, 2026-10-19).
“Whisper delivers state-of-the-art speech recognition with unmatched accuracy, enabling seamless voice interactions in our AI-powered websites and apps.”
LLM APIMaker says so · source ↗“We use variety of OpenAI models (o1, 4o, 4o-mini) and leverage their multi-modal and fine-tuning capabilities to enable multi-step categorization and other domain specific workflows.”
LLM APIMaker says so · source ↗“We use the Whisper models for speech analysis on device.”
LLM APIMaker says so · source ↗“We spent months developing the new version of Tellers.AI, betting that foundation models would excel at long-form, structured outputs with native tool use. GPT-5 did not disappoint; it proved to be the final missing cornerstone of our video-editing agent. The only bémol: api response delays and inference time.”
LLM APIMaker says so · source ↗“Regardless what the latest hype says, OpenAI models are insane!”
LLM APIMaker says so · source ↗“to power the AI agent that understands your network, enriches contacts, and surfaces high-value connections with precision.”
LLM APIMaker says so · source ↗“We’ve been using OpenAI since its early days. It has always been the easiest way to start building and commercializing LLM models, thanks to its user-friendly API and powerful capabilities”
LLM APIMaker says so · source ↗“We use GPT-5.6 to power the Basedash AI data analyst. It's incredibly intelligent and, with the right harness, allows us to rank #1 on BI Bench for solving real-world BI scenarios.”
LLM APIMaker says so · source ↗“We chose the GPT Image API because of its unmatched consistency. When you are building a product, you need a model that actually listens to what the user asks for without requiring them to become 'prompt engineers.' It reliably provides a high-quality, accurate starting point every single time, which is exactly what our users need to get their work done smoothly.”
LLM APIMaker says so · source ↗“Handy uses whisper as the model to do transcription, which enables it to run on most people's computers”
LLM APIMaker says so · source ↗
Gemini API
Model provider6 Astro products here (22%). Free up to about 360 users. About $28 a month at 10,000 users (Gemini 3.1 Flash-Lite).
In Lovable, ReplitBest forVery long context and multimodal input, with a free tier to start.
Trade-offFree-tier prompts may be used to improve Google's products, so paid tier is the one for user data.
“Saturn uses Gemini for structured data extraction from Japanese government filings (EDINET, gBizINFO). Best cost-performance ratio for Japanese language processing at scale.”
- It offers some of the best price-to-performance, with fast, cheap Flash models. 4 sources
- A context window of a million tokens or more handles whole codebases and long documents without extra pipelines. 4 sources
- Getting an API key and paying is confusing, with usage tiers and Google Cloud console hoops. 4 sources
- Models are deprecated abruptly, some without leaving preview or having a replacement. 4 sources
“Kuku uses Gemini for its tool-calling capabilities, which enables the AI to go beyond chat. Gemini allows Kuku’s AI agent to search, read, and edit real files reliably with low latency and a generous free tier.”
LLM APIMaker says so · source ↗“Saturn uses Gemini for structured data extraction from Japanese government filings (EDINET, gBizINFO). Best cost-performance ratio for Japanese language processing at scale.”
LLM APIMaker says so · source ↗“The model that does the actual work. Gemini 3 Pro generates every template EmailTemple ships — full HTML, ESP-native, brand-aware. Tested every major frontier model against side-by-side outputs. Gemini won on structural soundness and instruction-following inside long system prompts. Long context made the multi-document architect prompts possible.”
LLM APIMaker says so · source ↗“Multimodal models for responsive AI experiences Gemini helps support fast, capable, and multimodal interactions across the Agentplace platform.”
LLM APIMaker says so · source ↗“for image generation. nano banana gets text in images right, which is still rare. fast and cheap to iterate on too.”
LLM APIMaker says so · source ↗“Killer model for our purposes. Affordable, strong performance, thanks Google!”
LLM APIMaker says so · source ↗
Claude
Model provider4 Astro products here (15%). Free up to about 1,000 users. About $10 a month at 10,000 users (Claude Haiku 5.5).
llms.txtIn Lovable, ReplitBest forCoding, long documents and agent-style tool use.
Trade-offNo image generation or embedding models of its own, so those need a second provider.
“A family of foundational AI models Claude Sonnet and Opus help us power reliable, high-quality reasoning across Agentplace”
- It is strong at writing code and powers many code generation and coding agent products. 4 sources
- It stays coherent through long, multi-step reasoning and agent workflows. 4 sources
- Outages are frequent, with uptime below three nines across the API and apps. 4 sources
- API pricing is a large multiple of subscription pricing, making heavy agent use expensive. 3 sources
Status page: 36+ incidents in the last 90 days, 17 major (its own page, 2026-10-19).
“I've been a Claude fanboy for a while, so it was the first LLM I went to when building Hedy.”
LLM APIMaker says so · source ↗“A family of foundational AI models Claude Sonnet and Opus help us power reliable, high-quality reasoning across Agentplace”
LLM APIMaker says so · source ↗
Groq
Open models, hosted2 Astro products here (7%). Free up to about 1,500 users. About $6.75 a month at 10,000 users (GPT OSS 20B).
Best forVery low-latency inference on open models, for real-time features.
Trade-offOnly the open models it chooses to host, and no frontier closed models.
“We use Groq for ultra-fast inference when analyzing millions of contact records and enriching them with AI. It enables us to run deep research and structured reasoning at speeds that would be impossible on standard GPU setups. This level of performance lets us deliver intelligent outputs in real time, even at scale. We're grateful to the Groq team for building the kind of infrastructure that makes this possible.”
- Inference is extremely fast, enabling real-time voice and agent workflows. 4 sources
- The free tier is generous, with many models to choose from. 3 sources
“We use Groq for ultra-fast inference when analyzing millions of contact records and enriching them with AI. It enables us to run deep research and structured reasoning at speeds that would be impossible on standard GPU setups. This level of performance lets us deliver intelligent outputs in real time, even at scale. We're grateful to the Groq team for building the kind of infrastructure that makes this possible.”
LLM APIMaker says so · source ↗“Great models, very accurate, very fast”
LLM APIMaker says so · source ↗
Grok API
Model provider2 Astro products here (7%). Free up to about 110 users.
llms.txtIn v0Best forGrok models with built-in web and X search, plus image, video and voice APIs.
Trade-offReal-time data only arrives when you enable the search tools, which are billed as tool calls.
“Fast models for real-time reasoning Grok adds fast, context-aware reasoning that helps expand the range of workflows our agents can handle.”
“Best cost-to-quality ratio for structured JSON generation. Chinilla needs every AI response in strict JSON (components, connections, behaviors) and Grok handles that reliably at a fraction of the cost per token. When you're burning hundreds of AI calls per user session on system design, code mapping, and simulation analysis, the pricing difference adds up fast.”
LLM APIMaker says so · source ↗“Fast models for real-time reasoning Grok adds fast, context-aware reasoning that helps expand the range of workflows our agents can handle.”
LLM APIMaker says so · source ↗
OpenRouter
Gateway (many providers, one API)2 Astro products here (7%).
llms.txtIn ReplitBest forTrying many models from many providers with one API key and one bill, without opening an account at each.
Trade-offA fee on top of provider prices, and your traffic and data pass through a third party.
“We didn’t want to commit to a single model. OpenRouter lets us route across providers, experiment quickly, and use the best model for each task without extra overhead.”
- One API key and OpenAI-compatible endpoint reach hundreds of models from many providers. 1 source
- Switching and comparing models is easy, which speeds up testing cost and quality. 1 source
- Quality varies by underlying provider, so users pin trusted providers to avoid degraded output. 3 sources
- Caching is less effective because requests can switch providers mid-session. 3 sources
“Openrouter is the quickest option to get support for any model”
LLM APIMaker says so · source ↗“We didn’t want to commit to a single model. OpenRouter lets us route across providers, experiment quickly, and use the best model for each task without extra overhead.”
LLM APIMaker says so · source ↗Also chosen, and not seen yet
Fewer Astro products here: Mistral AI (1), LiteLLM (1).
Real options for LLM API not seen among these Astro products yet: DeepSeek.
Who switches, and to what
Public GitHub pull requests whose title says "X to Y", by developers in general, not only on Astro.
- OpenAI API → Gemini API 178+ pull requests · how they moved
- Gemini API → Groq 152 pull requests · how they moved
- OpenAI API → Claude 127 pull requests · how they moved
- Gemini API → Claude 108+ pull requests · how they moved
- Claude → Gemini API 104+ pull requests · how they moved
- Gemini API → OpenAI API 90+ pull requests · how they moved
Getting LLM API right
Put model calls behind one function in your code so switching providers is a one-line change. Log prompts, responses and cost from day one. Set spending limits on every provider account.
- Shipping with no spending limit or per-user rate limit, so one abusive user or a retry loop runs up the bill overnight.
- Hard-coding one model name across the codebase, so moving to a newer or cheaper model means a search-and-replace and a round of surprises.
- Putting the provider API key in client-side code, where anyone can copy it from the browser.
The rest of the Astro stack
What makers' Astro products chose for their other decisions, the most common pick for each:
- Hosting: Vercel 49 of 112
- Transactional Email: Resend 44 of 84
- Web Analytics: Cloudflare Web Analytics 18 of 47
- Payments: Stripe 22 of 34
- Database: Supabase 7 of 27
- Product Analytics: PostHog 13 of 14
- Authentication: Clerk 8 of 12
- Speech-to-Text & Text-to-Speech: ElevenLabs 7 of 12
Questions
What do most Astro apps use for LLM API?
Among 27 makers' products built with Astro that we know the LLM API of, OpenAI API leads: 17 of them (63%). Next come Gemini API (6) and Claude (4).
What does LLM API cost for an Astro app with 10,000 users?
On our usage model at 10,000 users, OpenAI API comes to about $10 a month (GPT-6 Luna); Gemini API comes to about $28 a month (Gemini 3.1 Flash-Lite); Claude comes to about $10 a month (Claude Haiku 5.5); Groq comes to about $6.75 a month (GPT OSS 20B). Prices are from each vendor's pricing page.
Which LLM API option stays free longest for an Astro app?
Mistral AI: on our usage model its cheapest plan stays free up to about 1,700 users.
How this is counted: a product is on Astro when a live-product source says so (the maker, its website, a customer story or a subprocessor list), and the same for its LLM API. How evidence is collected.