Home → Guides
Written from things we hit running this platform. Every technical claim here was verified on our own infrastructure or is quoted from a vendor's published documentation.
List prices for the current Claude models, how cache reads and cache writes change the real bill, the Fable 5.1 parameters that return 400,
Opus 5.5 lists at $4 input and $20 output per million tokens against Opus 5's $5 and $25 — but the larger change is the cache read rate, whi
We ran the same three checkable tasks through gpt-6.1-sol and claude-opus-5-5 on this gateway. Both got every answer right. Opus wrote up to
gpt-6-sol lists at $2/$10 per million tokens, gpt-6-astra at $10/$50 — five times apart in the same family. Both reprice above 272k input to
grok-4.7 costs the same as grok-4.6 at list: $2 input and $6 output per million tokens below 200k prompt tokens, double above it. What that
A gateway puts several model providers behind one endpoint and one bill. Here is what that buys you, what it costs you, and the cases where
An OpenAI-compatible API accepts the same request shape as OpenAI's, so the official SDKs work unchanged. Here is exactly what carries over,
A practical migration guide: the two lines that change, the model-string mapping, what OpenRouter does that APICLAN does not, and how to run
How to connect Chatbox to an OpenAI-compatible API: the API Host / API Path combination that causes 404s, the different /v1 rule for Claude, why the I
How to add an OpenAI-compatible provider to Cherry Studio: the API address rule that causes most 404s, fetching the model list, adding Claude as an An
Claude Code reads its API address from ANTHROPIC_BASE_URL. The value must be a bare host with no /v1. Exact settings for macOS, Linux, Windows and set
Claude Fable 5 and Fable 5.1 both list at $10 input and $50 output per million tokens, but Fable 5.1 bills cache reads at a quarter of the rate. Real
How to connect Cline in VS Code to an OpenAI-compatible gateway: which API provider to pick, the exact base URL for each, which key to use, and our te
How to point Codex at a third-party OpenAI-compatible API: the exact config.toml block, where the file lives on macOS, Linux and Windows, how to set t
gemini-4-argon lists at $2 input and $10 output per million tokens until November 30, 2026, then doubles to $4 / $20. Cached input is 95% off. What th
gpt-6-astra lists at $10 input and $50 output per million tokens, $1 for cached input, and doubles to $20 / $75 above 272,000 input tokens. Here is wh
We gave GPT-6 Astra, GPT-6.1 Sol and Claude Opus 5.5 the same 13 JavaScript tasks twice and ran every answer against hidden tests in an offline sandbo
How much the ChatGPT / OpenAI API costs in 2026: per-million prices for gpt-6-astra, gpt-6.1-sol, gpt-6-sol, gpt-5.6-sol and gpt-5.6-terra, what a typ
How to connect OpenCode to an OpenAI-compatible API with opencode.json, why baseURL must end in /v1, how to add Claude, and our test of OpenCode 1.19.
Point the Claude desktop app and Codex at APICLAN with CC Switch: the exact request address to enter, why one client needs /v1 and another does not, a
One page per model, built from this platform’s own billing records: what calls actually cost here, how much of that went to cache reads versus cache writes, measured latency, and the workloads where the model is the wrong choice. The numbers regenerate from the database, so they do not go stale.
claude-fable-5-1 bills cache reads at $0.25 per 1M — a quarter of what every other Claude model charges. Verified against a real invoice, with the cac
A real claude-opus-5 invoice where 90 input tokens and 8,338 output tokens still cost $1.13 — because 81% of it was cache writes. The arithmetic, why
What claude-sonnet-5 actually costs, measured over 130 billed calls: median vs mean per call, why they differ by 3x, how much of the bill is cache wri
What gpt-5.6-sol actually costs across 29 billed calls, why its first-token time was the slowest of the GPT models we measured, and when terra or luna
What gpt-5.6-terra actually costs across real billed calls, how it compares with sol and luna on measured cost and latency, and the workloads where it
What gpt-6-astra actually costs from 14 real billed calls, how we verified the list price against an invoice, the 272,000-token threshold that reprice
The model-name suffix sets the resolution, not the size parameter. Measured delivery for every tier, the 1.57 MP cap on the plain name, and flat per-i
If you are wiring up a coding agent right now, go straight to base URL setup — it has the exact setting for each client and the one mistake that produces an error you cannot find in any log.
If you are evaluating whether to use a relay at all, what an LLM gateway is lays out what it buys you and what it costs you, including the cases where calling the provider directly is the better engineering decision.
If you are trying to understand a bill that came in higher than expected, Claude API pricing in 2026 covers cache reads versus cache writes, which explains more Claude invoices than the headline per-token rate does.
There are no capability comparisons here and no benchmark tables. We have no evaluation data of our own, and repeating someone else's would not predict how a model behaves on your task. Where a decision depends on model quality, the honest answer is to run two candidates on twenty real inputs from your own product and read the outputs. That takes an afternoon and settles it better than any page we could write.
You will also not find uptime figures, concurrency limits or customer counts, for the same reason: we would not be able to back them up.
No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.
Read the 30-second quickstartPrices quoted on this page are regenerated automatically from live billing data. Third-party terms are quoted from that party's own published documentation.