⚠️
gpt-image-2-vip upstream org-level rate limit alert —— Recent errors like Rate limit reached for gpt-image-2-codex ... in organization ... on input-images per min: Limit 4000, Used 4000 come from OpenAI’s official-side org-level rate limiting: OpenAI links multiple accounts under the same organization, so all users in that org share a 4,000 input-images-per-minute quota — once it’s hit, requests are throttled. This is largely unrelated to our account pool — it’s an upstream issue. If it affects your business, we recommend switching to the official-relay gpt-image-2: the call method (images API) and parameters like Size are fully compatible.📖 View details🚀
MiniMax-M3 Launch · 50% off matching official, down to ~41% of list —— MiniMax’s new open-weight flagship MiniMax-M3 (case-sensitive) is live on APIYI! 1M-token context + native multimodality, SWE-Bench Pro 59.0 beating GPT-5.5. Limited-time 50%-off rate: $0.30 input / $1.20 output per 1M tokens (0-512K tier), down to ~41% of list with recharge bonuses, ending June 8, 00:00 (UTC+8) — subsequent pricing TBD. 📖 Deep dive📖 View details✅
gpt-image-2-vip recovered · error handling changed: 500 system errors disclosed directly, no retry loop —— Around 6 PM (UTC+8) this evening, gpt-image-2-vip is back to normal. We also adjusted the error handling: when an An error occurred while processing your request. type 500 OpenAI system error occurs (often on 4K or compute-heavy complex image requests), we now disclose the error directly instead of retrying repeatedly and dragging out the timeout — so you get clear feedback faster.📖 View details💰 Big DeepSeek price cut ·
deepseek-v4-pro / deepseek-v4-flash group now 5% off —— We’ve synced the latest official pricing — the DeepSeek series gets a big price cut! The deepseek group multiplier is lowered to 0.95x (i.e. 5% off), and it stacks with top-up promotion discounts for an even lower price. China’s pride — hats off to the DeepSeek team. A value-packed model, enjoy! This adjustment covers deepseek-v4-pro (input $0.43 / output $0.86) and deepseek-v4-flash (input $0.14 / output $0.28), per 1M tokens.📖 View details⚠️ codex risk control tightens again ·
gpt-image-2-vip expected to be unstable short-term —— codex has rolled out heavy new risk control: ① the account pool we topped up last night lost its authorization status across the board today; ② on top of phone verification, a second round of 2FA was added. As a result, gpt-image-2-vip is expected to be unstable in the short term. Status of other channels right now: official-relay gpt-image-2 is running stably, gpt-image-2-all is running stably; image needs can also use the cheaper gemini-3.1-flash-image-preview (Nano Banana 2).📖 View details💰
claude-opus-4-8 pricing corrected · aligned to official $5 / $25 —— Since May 29, due to a pricing misconfiguration, claude-opus-4-8 had its output price set to $15 (it should be $25), so it was billed at the lower rate for 5 consecutive days. This was a platform-side issue — we absorb the shortfall and won’t claw it back. It has now been corrected to $5 input / $25 output per 1M tokens, aligned with the official standard. Thanks to Archer for the feedback. We’d also emphasize: we run pure AWS Bedrock official-relay Claude — reliable quality, with a cache hit rate of basically 100% (for eligible requests) — a comfortable experience. Thanks to all our customers for their understanding.📖 View details✅
gpt-image-2-vip back to normal · recommend wiring up official-relay gpt-image-2 as backup —— Following up on the daytime risk-control alert: gpt-image-2-vip is back to normal. That said, the reverse-proxy channel may still be unstable at times in the future, so we recommend also wiring up the official-relay gpt-image-2 as a backup, just in case. For setup and usage, see the documentation center: gpt-image-2 docs.📖 View details⚠️ Strict codex risk control ·
gpt-image-2-vip affected, no firm recovery ETA —— Today codex risk control is very strict and the codex quota has been exhausted, so gpt-image-2-vip is affected and we are adding resources; we cannot give a firm recovery ETA, because newly added accounts still get flagged by risk control. Alternatives: ① official-relay gpt-image-2 produces 4K with controllable quality; ② gpt-image-2-all works for general needs; ③ the Nano Banana series is mostly fine but occasionally needs a retry and is currently in an unstable state.📖 View details⚠️
gpt-image-2-vip quota exhausted under high concurrency · errors appearing, adding resources —— Follow-up to this morning’s alert: as noted, gpt-image-2-vip is a premium-quota reverse-proxy model, and its codex quota is being run down under high concurrency, so some errors are now appearing — we are adding resources. The error message advises “contact support if errors are frequent.” In the meantime you can switch to the official-relay gpt-image-2, currently the most stable — callable on the standard group and steadier on the enterprise group.📖 View details⚠️
gpt-image-2-vip (reverse-proxy) occasional concurrency shortfall · adding resources —— Due to codex tightening phone-number verification, the reverse-proxy channel gpt-image-2-vip occasionally has insufficient concurrency, and we are adding resources. If you hit 429 errors, switch to the official-relay gpt-image-2; the official-relay gpt-image-2 is currently the most stable — callable on the standard group and even steadier on the enterprise group.📖 View details📖 For earlier updates, visit the Live Updates Archive — browse history by month, category, or vendor.
Deep Dive
AI Radar
Telegram
Global users
Media Models
New additions