OpenAI and Claude API access in Thailand, minus the billing maze.

Quick answer: base URL, key, exact model id.

Point your tool or SDK at https://api.apisrouter.com/v1 with an APIsRouter key and request models by id: gpt-5.5, claude-sonnet-4-6, gemini-3.5-flash, deepseek-v4-flash. The wire format is the standard OpenAI Chat Completions dialect, so nothing about your code or tooling is exotic; if it takes a custom base URL, it works. The billing model is the part built for this situation: a prepaid balance topped up when you choose, no card on file, and a free start so the first tests cost nothing. Keys arrive by email, and each key's usage is visible per model in the console.

curl https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.5",
       "messages":[{"role":"user","content":"สวัสดี"}]}'

What OpenAI and Anthropic actually support in Thailand.

Verified as of July 2026: Thailand appears on OpenAI's supported countries list for the API and on Anthropic's supported regions list. Google's Gemini API is available too, with a free tier in AI Studio that requires no card. Account creation, console access, and API usage from Thailand are all within the vendors' terms. Both vendors revise these lists, so treat their official pages as the source of truth. But the pattern for Thai developers has been stable for a while: signing up works, and the setup stalls at the payment form, where the vendor wants an internationally enabled card and Thailand's daily payment rails do not qualify.

Why Thai payment methods stall at vendor checkouts.

Because AI APIs bill continuously rather than once, an attached card is a recurring liability: each charge is a fresh chance for the issuer to decline, and a declined renewal can pause a production workload. A prepaid balance fails differently and more predictably: it runs out visibly, and topping up is a deliberate act rather than a background charge.

• PromptPay and TrueMoney are domestic rails: they settle in THB inside Thailand and cannot attach to a US vendor's USD billing account.
• Cross-border card blocks: Thai banks commonly ship debit cards with international online payments disabled, and enabling them is a per-bank, sometimes per-card-type process that not every account supports.
• Debit-heavy card market: many Thai developers hold debit rather than credit cards, and US billers decline foreign debit cards at higher rates, with generic error messages when they do.
• Prepaid cards rejected: OpenAI explicitly refuses prepaid cards, closing the most obvious workaround.
• 3D Secure and address checks: verification redirects and billing-address validation both fail more often on cards issued outside the biller's home market.

One endpoint, every major model family.

APIsRouter is a gateway in front of the major vendors: one base URL, one key, per-request routing by model id. From Thailand that means a single top-up covers GPT, Claude, Gemini, DeepSeek, GLM, Kimi, and Qwen ids, and switching vendors is an edit to a string rather than a new billing fight. Everything that speaks the OpenAI dialect plugs in: the official SDKs, coding agents like Cline and Aider, chat clients like Chatbox and LobeChat, and self-hosted stacks like Dify, FastGPT, and Open WebUI. For teams, per-key usage means each project or developer gets its own key and its own line in the usage view.

Latency from Thailand: the honest version.

APIsRouter serves from cloud regions in the EU and US, not from Bangkok or Singapore. Every request from Thailand crosses continents first, which adds a real fraction of a second of round-trip time before generation begins. Whether that matters depends entirely on the workload. Coding agents, document pipelines, batch classification, and anything streamed absorb the overhead easily, because model generation time dominates and streaming delivers tokens as they come. A realtime, latency-critical chat product for Thai consumers is the case where a vendor's regional infrastructure wins where you can get its billing working. Do not guess: time a streamed request from your deployment region, read the time-to-first-token, and compare it against what your product actually needs.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

A sensible model mix for Thai workloads.

All five ids bill through one key, and the console attributes tokens and spend per model. Run your real traffic across two candidates for a week and let the usage view decide; that beats any static comparison table, including this one.

• gpt-5.5 as the frontier generalist, the natural first pick for anyone arriving from the OpenAI ecosystem.
• gpt-5.4-mini for volume work in the GPT family: drafts, tagging, extraction, and support flows.
• claude-sonnet-4-6 for coding agents and tasks where instruction-following quality is the bottleneck.
• deepseek-v4-flash as the floor-rate volume tier for Thai-English translation passes and high-message products.
• gemini-3.5-flash for long-context and image-input workloads at a mid-tier rate.

Verify the endpoint, then wire in your tools.

List the models first: it proves the key works and gives you the exact id strings. Then send one completion with the id you plan to standardize on. After those two commands succeed, any remaining failure lives in your tool's settings rather than in the gateway. First-run failures are predictable: 401 means the key is not on the request, model-not-found means the id string is off by a character, and universal 404s mean the base URL is missing /v1. From there, the console's per-request log of model, tokens, and spend turns day one into a realistic monthly budget number.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | head -50

FAQ