Skip to main content
Welcome to the APIYI changelog. This is where we announce new model launches, pricing changes, and feature updates. We’re committed to giving you the most capable AI services at the best value.
Bookmark this page: press Ctrl+D to keep it handy — new model launches and promotions are announced here first.

​
🔥 Latest Updates

​
New Model · Zhipu

​
GLM-5.3 and GLM-5.3-Flash Are Live

Zhipu’s Coding Flagship and Multimodal Lite, at Official PriceZhipu’s two August releases arrive together: glm-5.3 keeps the 753B MoE base and scales post-training only, gaining 50% over 5.2 on Z.ai Code Bench; glm-5.3-flash is the first natively multimodal GLM-5, scoring 84.3 on Terminal-Bench 2.1, within reach of Claude Opus 4.8. Priced item for item with Zhipu’s official rates — the flagship at $1.40 in / $4.396 out, Flash at $0.15 in / $0.50 out per 1M tokens — in the default / svip groups, with recharge bonuses bringing the effective cost to roughly 83%–91% of list.📖 Read details | 🔗 Promotions
​
New Model · OpenAI

​
GPT-6 Astra Is Live

OpenAI’s New Flagship for Computer Use and Long-Horizon Agents, at Official PriceOpenAI’s next-gen flagship, released 3 September: 1.05M context, five reasoning-effort levels with new xhigh / max, Terminal-Bench 4.0 up from 37.3% to 57.9%. Priced at $10 in / $50 out per 1M tokens with $1 cached reads, matching OpenAI’s list price item for item, in the default / svip official-relay groups, with the Codex_Reverse group at a 0.5x discount.📖 Read details | 🔗 Promotions
​
Price Update · OpenAI

​
GPT-5.6 Sol Price Cut Synced

Input Down 20%, Output Down 33%, Now Cheaper Than gpt-5.5OpenAI cut GPT-5.6 Sol pricing on 3 September and APIYI has synced: gpt-5.6-sol now costs $4 input (was $5) and $20 output (was $30) per 1M tokens, cached reads $0.40, with the limited-time promotional rate guaranteed at least through 21 November 2026. The flagship tier is now cheaper than the previous-generation gpt-5.5 ($5 / $30); existing code migrates by changing only the model field.📖 Read details | 🔗 Model details
​
New Model · Google

​
Gemini 3.8 Flash Is Live

Ahead of the Docs, Same Price as 3.7Google’s newest Flash, out 2 September. Google’s own model docs and launch blog have not listed it yet; APIYI has it open for calls. Priced line-for-line with gemini-3.7-flash at $0.75 in / $3.75 out per 1M tokens, so migrating costs nothing. 150 paired test cases across both protocols found capability parity with 3.7 and no regression unique to it.📖 Read details | 🔗 Promotions
​
New Model · Anthropic

​
Claude Fable 5.1 Is Live

Same Price, Cache Reads Cut to a QuarterAnthropic’s new Mythos-class flagship, shipped 1 September. The headline change is cache reads falling from $1.00 to $0.25 per 1M tokens, with input at $10 and output at $50 unchanged. APIYI has matched the cut — every line item is priced in line with the provider. Three breaking changes: forced tool use now 400s, thinking blocks are model-bound, and editing earlier turns invalidates them.📖 Read details | 🔗 Promotions
​
Price Update · ByteDance

​
Seedance 2.5 and the 2.0 Family Share One Group

A Single Token Covers All Four ModelsAll four models — 2.5 plus the 2.0 standard / fast / mini — now sit on the SeeDance2 group (0.18x), reachable from a single token with no separate token for 2.5 and no code changes. Measured rates come in two tiers: $12.60 per million tokens with no video in the input (480p / 720p / 1080p alike), and a lower $7.56 when the input contains video (multi-modal reference, video editing / extension).📖 Read details | 🔗 Groups and pricing
​
New Model · ByteDance

​
Seedance 2.5 Is Live

30-Second Clips, 30 Reference Images, A Broad Capability JumpThe model is doubao-seedance-2-5-260628, with an endpoint and request shape identical to 2.0 — changing the model field is the whole migration. The duration cap rises from 15 to 30 seconds and reference images from 9 to 30, audio can stand alone as a reference, and it adds mov output plus explicit video edit/extend task types. For current group and pricing, see the 31 August entry above.📖 Read details | 🔗 Capabilities and pricing
​
New Model · DeepSeek

​
DeepSeek Vision Model Is Live

No Vision Premium, Priced Like Text-Only V4 Flashdeepseek-v4-flash-vision-exp is DeepSeek’s first vision model, adding image input on top of the V4 Flash base while keeping the 1M context, thinking, function calling and caching. Images convert to input tokens by size and are capped at 384 each; pricing matches the text-only version at $0.44 input and $1.32 output per 1M tokens. Use a default group token for the OpenAI format and a ClaudeCode one for the Anthropic format.📖 Read details | 🔗 Integration docs
​
Feature Update · OpenAI

​
Transparent Backgrounds on gpt-image-2

One Parameter, Real Alpha-Channel PNGsOpenAI opened up the transparent value of background for GPT-Image-2, and APIYI has verified it. Pass background: "transparent" with png or webp to get a genuinely transparent image; text-to-image, image editing, and the Responses image tool all support it, at no extra cost. jpeg has no alpha channel and is mutually exclusive with transparency.📖 Read details | 🔗 Transparency FAQ
​
Price Update · DeepSeek

​
DeepSeek Price Change

Peak/Off-Peak Upstream, Peak Tier at APIYIDeepSeek switches to two-tier billing at 00:00 on 17 August (UTC+8), with off-peak at half the peak rate. APIYI follows for deepseek-v4-flash and deepseek-v4-pro but bills at the peak tier at all times: $0.44/$1.32 and $1.32/$3.96 per 1M tokens. The reason is supply — official capacity shortfalls are covered by the pricier BytePlus and Alibaba Cloud routes, at no margin. Recharge bonuses still stack.📖 Read details | 🔗 Promotions
📖 These are the 10 most recent announcements. For everything earlier, see the announcement archive, browsable by month, category, or vendor.

Deep dives

Browse the AI Radar section

Live updates

Model status and service notices

Promotions

See current deposit bonuses