Models & Pricing

Models & Pricing

The same models as Anthropic and OpenAI direct. The difference: pay per token with QRIS. No USD card, no monthly subscription, no FX card fee.

Display name vs Model ID

Names such as GPT-5.6 Sol are human-readable display names. When calling the API, use the exact Model ID, such as gpt-5.6-sol. Model IDs are case-sensitive: do not change capitalization or pass the display name as the model value.

List every Model ID directly from the live catalog:

curl -s https://api.nexotao.com/models | jq -r '.models[].model'

All 12 live models and their Model IDs:

Display nameAPI Model ID
DeepSeek V4 FlashDeepSeek-V4-Flash
DeepSeek V4 ProDeepSeek-V4-Pro
GPT-5 Minigpt-5-mini
GPT-5.6 Lunagpt-5.6-luna
GPT-5.6 Solgpt-5.6-sol
GPT-5.6 Terragpt-5.6-terra
Grok 4.3grok-4.3
Claude Opus 4.6claude-opus-4-6
Claude Opus 4.7claude-opus-4-7
Claude Opus 4.8claude-opus-4-8
Claude Opus 5claude-opus-5
Claude Sonnet 4.6claude-sonnet-4-6

The live catalog is what counts — if this table and GET /models disagree, trust GET /models.

Choosing a model

What each model is for. For its rates, see the live table below.

ModelBest for
DeepSeek-V4-FlashHigh-volume, cheap work: classification, extraction, short chat.
DeepSeek-V4-ProHigher-quality DeepSeek; 131,072-token context.
gpt-5-miniOpenAI’s cheapest tier — classifiers, extraction, batch. Supports vision.
gpt-5.6-lunaEfficient for everyday work and high-volume tasks.
gpt-5.6-terraBalanced for coding, analysis, and agent workflows.
gpt-5.6-solTop GPT-5.6 tier; the same rate for input and output.
grok-4.3An alternative outside the GPT and Claude families.
claude-opus-4-6The Opus that still accepts temperature / top_p / top_k.
claude-opus-4-7Previous-generation Opus; sampling parameters no longer accepted.
claude-opus-4-8The strongest reasoning model for coding and agentic work.
claude-opus-5The newest Opus; 1,000,000-token context.
claude-sonnet-4-6Fast + smart for high-volume product chat; sampling still works.

The Claude models and gpt-5-mini support vision (image input); the rest are text-only. Check supports_vision in GET /models to confirm.

Match the model to the job. DeepSeek is great for high-volume, cheap workloads — classification, extraction, short chat — but not for agentic/coding or large context. Its capacity is small (~20k tokens/min), so a single agentic or large-context request can hit 429 server at capacity. For coding/agentic & large context, use Claude (Opus/Sonnet) or GPT-5.6.

Live per-million-token rates

Every model has per-million-token rates for input, output, and (for Claude/GPT) cache-read + cache-write. The values here come straight from GET /models — always in sync with the live catalog, with no hardcoded prices on the page.

Loading live pricing…

How to read the rates

  • Text — rate per 1 million tokens, separately for input and output.
  • Cache (Claude/GPT) — cache reads are much cheaper than the standard input rate; cache writes are slightly more. Full breakdown in Billing & Pricing.
  • What you’re billed — always token_count × model_rate_at_the_time, rounded up. There is no per-request fee and no minimum charge.

Need a cost estimate before you call the API? Take the rate from GET /models and multiply it by your expected token count — or see nexotao.com/harga for a calculator and price comparisons.

Billing details, cache mechanics, and per-request accounting live in Billing & Pricing.