What we ship.

GLM-5.3 Flash — live

GLM-5.3 Flash is live. Z.ai's 320B-parameter MoE (18B active) — natively multimodal, 1.28M-token context, approaching Claude Opus 4.8 on coding and agentic benchmarks — at $0.15 / $0.50 per 1M tokens. Through September 30, top up $5-$99 and your credit is tripled ($5 → $15, $50 → $150, up to $100 of bonus per account); the bonus is spendable on GLM-5.3 Flash only and expires 90 days after the top-up. A single top-up of $100 or more instead makes GLM-5.3 Flash unmetered for the same window.

GLM-5.3 Flash is live on AIgateway — Z.ai's 320B-parameter MoE with 18B active parameters, natively multimodal, with a 1.28M-token context. It approaches Claude Opus 4.8 on coding and agentic benchmarks at $0.15 / $0.50 per 1M tokens. Launch offer, in two tiers split at $100: top up $5-$99 through September 30 and your credit is tripled — $5 holds $15, $50 holds $150 — with the bonus spendable on GLM-5.3 Flash only, expiring 90 days after the top-up, capped at $100 of bonus per account. A single top-up of $100 or more takes the other tier: no bonus, but GLM-5.3 Flash stops being metered for you through the same date, while the $100 stays ordinary credit for the whole catalog. GLM-5.3 Flash also headlines the free trial catalog: add a card (no charge) and get $1 of trial credit to start. Model page: /models/zai-org/glm-5.3-flash.

Free GLM-5.2 — no card needed

GLM-5.2 — a Claude-class coding model — is now free to use with no card, on fair-use limits. Add a card (no charge) for $5 across a curated trial catalog, or top up any amount for the full 1000+ catalog.

GLM-5.2, a Claude-class coding model, is now free to use with no card on fair-use limits — sign in, grab an `sk-aig-…` key, and start building. Want more? Add a card (no charge) to unlock $5 in credit across a curated trial catalog (Kimi K2.7 Code, Kimi K2.6, GLM-5.2, BGE-M3, FLUX-2 Klein, Gemma 4, Aura 2, Whisper Turbo, Llama Guard); the $5 expires 30 days after you add a card. Top up any amount to open the full 1000+ catalog at pass-through + a 5% platform fee — topups never expire.

Auto Router

Auto Router is live. Set model:"auto" (or scope it with model:"auto/text", auto/image, auto/video, auto/tts, auto/stt, auto/music, auto/embedding) and AIgateway reads each request, picks the cheapest model in a curated, eval-covered pool that still clears the quality floor, and bills you less than the premium model you'd otherwise have called — guaranteed, with the baseline acting as a hard cost ceiling. It's the only auto router that spans every generative modality, not just text. Every routed response returns transparency headers showing the model that ran, why, your premium baseline, and the exact dollars saved. Optional baseline_model sets your own ceiling; x-routing (cost | speed | quality | auto) biases the pick.

Set `model:"auto"` on any request and the router reads it, picks the cheapest model in a curated, eval-covered pool that still clears the quality floor, and never charges you more than the premium model you'd have called yourself. Scope a lane with `model:"auto/text"` (also image, video, tts, stt, music, embedding) or omit the model field entirely. Every request carries a baseline — set it with `baseline_model`, or it defaults to the premium model for that modality — and the router only routes down from it, so the baseline doubles as a hard cost ceiling. Every routed response returns transparency headers: X-Routing-Selected, X-Routing-Reason, X-Routing-Complexity, X-Routing-Quality, X-Auto-Baseline-Model, X-Auto-Baseline-Cost-Cents, and X-Auto-Savings-Cents. Bias the pick with the x-routing header (cost | speed | quality | auto). It's the only auto router that spans every generative modality, not just text. Full walkthrough on /auto-router and /docs/auto-router.

/v1/images/edits — OpenAI-compatible image editing

client.images.edit() now works against the gateway. Multipart upload or image_url JSON, optional mask, default model bria/fibo-edit/edit. Backed by the full Bria Fibo edit family, Bytedance Seedream V4 edit, and Flux image-to-image — pick by model id.

The endpoint was previously a 501 stub. It now accepts multipart/form-data (image File + optional mask File + prompt) or JSON ({ image_url, mask_url, prompt }). Multipart uploads are content-addressed and archived to R2, so we hand fal a stable media.aigateway.sh URL instead of stuffing base64 into the upstream JSON body. 25 MB inbound cap per file. Response shape matches OpenAI exactly: { created, data: [{ url | b64_json }] }. Default model when none is specified is bria/fibo-edit/edit (~$0.04/image, general-purpose). Override with any model whose capabilities include image-to-image — Bria Fibo erase/replace/restyle/relight/colorize, Bytedance Seedream V4 edit, Flux dev, Bria Upscale Creative, and more.

Python SDK, Node SDK, and CLI — live on PyPI + npm

pip install aigateway-py · pnpm add aigateway-js · npm i -g aigateway-cli (`aig` binary). Wraps async jobs, sub-accounts, evals, replays, signed file URLs, and webhook verification.

The OpenAI SDK with our base_url is still the right choice for chat / embeddings / images / STT / TTS — that hasn't changed. The new packages cover the aggregator-native surface OpenAI doesn't model: text-to-video / music / 3D async jobs with a typed wait() helper, per-customer sub-account keys, eval-driven routing, replay + shadow A/B, signed file URLs, and HMAC webhook verification. The CLI ships as the `aig` binary — `aig init` walks through key entry, writes .env, and scaffolds a working starter file in TS / JS / Python. Source on GitHub at packages/{sdk-python,sdk-node,cli}.

AIgateway is live

One OpenAI-compatible API to 1000+ models across 85+ labs. Every modality. $5 signup credit on a curated 7-model edge tier.

Today we open the doors. One API key, every frontier model, every modality — text, image, video, voice, audio, embeddings — behind one schema. Drop-in for the OpenAI SDK; only the base_url changes. Every new account gets $5 in signup credit redeemable on a curated edge tier (Kimi K2.6, BGE-M3, FLUX-2 Klein, Gemma 4 vision, Aura 2, Whisper Turbo, Llama Guard); the credit expires 7 days after signup.

Per-route Open Graph cards

Pricing, models, agents, and enterprise pages each render their own dynamic OG image. Better link previews everywhere.

Public brand kit

Mark, wordmark, lockup in SVG and PNG at every size, dark + light variants, plus avatar / Twitter header / GitHub social PNGs. Live at /brand/.

Agent autoconfigure surfaces

Point Claude Code, Cursor, Cline, or any capable coding agent at /llms.txt, /agents.md, or /openapi.json — it configures itself.

Sub-account API

Mint scoped API keys per end user with spend caps, rate limits, default tags, and isolated analytics. One POST /v1/sub-accounts.