One Interface.
Every AI Model.

# Step 1: Agent gets its own Key (once)

→ API Key returned instantly, no human required

# Step 2: Start calling models for free

→ Smart routing to the best free model, zero cost

[Maintenance] Online card top-up temporarily unavailable

Due to payment system maintenance, online credit card top-up is temporarily unavailable. Your account balance and API service are NOT affected and continue to work normally. Top-up via bank wire transfer is also unaffected and remains available. If you urgently need to top up or have any questions, please contact us via in-site support. We apologize for the inconvenience and will announce the restoration time separately.

2026-07-07

Taiwan Payment & Invoice System Coming Soon

BazaarLink’s Taiwan payment and unified invoice system is currently under integration. Until then: The API is fully available New users can pay via Stripe and start testing immediately Beta feedback will help us improve the platform Developers are welcome to try BazaarLink today. Charges during the beta period will not include Taiwan unified invoices.

2026-03-10

[New] Bring Your Own Key (BYOK) and Content Filtering are live

Two new features are live today: ◆ Bring Your Own Key (BYOK): attach your own upstream provider API keys (OpenAI- or Anthropic-compatible) to your account or organization. Eligible requests go upstream through your key, with seamless or strict fallback modes. Personal setup lives under API Keys → BYOK; organizations manage theirs in org settings. ◆ Content Filtering: two-way protection for your API traffic — requests carrying prompt-injection attempts are blocked, and sensitive data in prompts or responses (API keys, card numbers, national IDs, and more) is automatically redacted. Rules and an exemption allowlist are customizable, with usage statistics included. Enable it on the Content Filter page. See the Quick Start section of the docs for details.

2026-07-28

Claude Opus 5 is live - #1 on Artificial Analysis, 1M context

Claude Opus 5 is now live. Call it with the model name claude-opus-5 (compatible alias: anthropic/claude-opus-5). Anthropic's current flagship model, built for demanding reasoning, coding, and long-horizon agentic work. It is particularly strong at end-to-end software tasks, code review and bug finding, and visual analysis. With a 1M-token context window it can take in an entire codebase or long document in one pass. [Third-party benchmarks] On the independent Artificial Analysis leaderboard: Intelligence Index 60.7 - ranked #1 of 573 models Coding Index 78.0 - ranked #2 of 198 models (Measured at Max Effort; data retrieved 2026-07-25) [Pricing (per 1M tokens)] Input: $5.00 Output: $25.00 Cached input: $0.50 (10% of the input price) Cache write: $6.25 [Specs] Context: 1,000,000 tokens Input: text, image, file -> Output: text Supports reasoning (tunable via reasoning_effort), tool calling, structured outputs, and verbosity control [Note] Opus 5 is a reasoning model and its responses are verbose. At Max Effort it generates around 44 tokens/sec with roughly 29s to first token - noticeably slower than average. If latency matters, lower reasoning_effort: High gives about 10s to first token (Intelligence Index 58.9) and Low about 2.8s (50.6), at the same price. For workloads with repeated prefixes (system prompts, long documents, codebases), use prompt caching - cached input drops to 10% of list price.

2026-07-25

Kimi K3 is live - 1M context, native vision

Kimi K3 is now live. Call it with the model name kimi-k3. Moonshot AI's most capable model to date: 2.8T parameters, native vision understanding, and a 1M-token context window, built for software engineering, knowledge work, and deep reasoning. It scores 57 on the Artificial Analysis Intelligence Index, placing it among the top tier of publicly available models. [Pricing (per 1M tokens)] Input: $3.00 Output: $15.00 Cached input: $0.30 (10% of the input price) [Specs] Context: 1,048,576 tokens Input: text, image -> Output: text Supports reasoning, tool calling, and structured outputs [Note] K3 is a reasoning model - responses are verbose and generation runs at roughly 62 tokens/sec, slower than average. Evaluate before using it in latency-sensitive paths, or consider the Kimi K2 series.

2026-07-17

新功能：auto:free — 免費智能路由模型

我們推出了全新的免費模型路由功能！只需將 model 設為 "auto:free"，系統會根據你的任務類型自動選擇最適合的免費模型，完全不扣費。 需要註冊帳號 + API Key（不需要儲值） 有 RPM 和每日請求上限 有儲值的用戶享有更高上限

2026-03-21

歡迎使用

一個平台滿足您所有 AI 推理需求

2026-02-25