NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #27 by repository stars
Latest stable
v0.13.2
Last release today
17 Nov 2026
Ships on a steady schedule
a new release about every 8 days
No release notes found
nothing matched a version
Nothing withdrawn
no release was ever pulled
1 years old
647 releases · first in 2025
Drag across the ruler to compare, or tap a release to open it.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
v1.0.0-rc.19-i18nfix.2 Compare # Choose a tag to compare
v1.0.0-rc.19-i18nfix.2
Compare
Nothing published for this version
Important 修复 v1.0.0-rc.50 在有 AES-NI、没有 AVX 的 x86-64 CPU 上崩溃并反复重启的问题( #8314 )。Celeron、Pentium、Atom 一类 CPU(如 J4125)和只提供 AES、不提供 AVX 的虚拟机会受影响:处理聊天等请求时进程因
Important
修复 v1.0.0-rc.50 在有 AES-NI、没有 AVX 的 x86-64 CPU 上崩溃并反复重启的问题(#8314)。Celeron、Pentium、Atom 一类 CPU(如 J4125)和只提供 AES、不提供 AVX 的虚拟机会受影响:处理聊天等请求时进程因 SIGILL 退出,模型列表和管理后台仍能打开,看起来像间歇性的 502。在这类机器上运行 rc.50 的请升级到本版本。
本版本包含 rc.50 新增的提供商插件、请求处理插件和社区插件市场。提供商插件把一个上游厂商接成一种渠道;请求处理插件在请求和回复上运行,可以按全局、分组、渠道和模型生效。从 rc.49 或更早版本升级时请注意下面几点。
原敏感词设置在首次启动时迁成一个「请求过滤」实例,旧设置项会被删除。命中时返回 400 request_rejected,原来是 500 sensitive_words_detected。回滚到旧版本后敏感词检查是关闭的,需要重新填写。
多节点部署:设置、权限、插件、渠道等手动改动几秒内同步到所有节点。CHANNEL_UPDATE_FREQUENCY 的定时余额更新改为只在主节点运行。超过 30 天的已结束系统任务会被自动删除,需要保留更久的请先在 系统设置 → 运维 → 日志维护 里改保留天数。
Vertex AI 插件 1.3.1 起,视频请求发到渠道填的部署地区,不再固定发到 global。升级前请确认所用的视频模型在该地区可用。
Fixes v1.0.0-rc.50 crashing and restarting over and over on x86-64 CPUs that have AES-NI but not AVX (#8314). Celeron, Pentium and Atom parts (such as the J4125) and virtual machines that offer AES but not AVX are affected: the process exits with SIGILL while handling chat and similar requests, while the model list and the admin console still open, so it looks like an intermittent 502. If you run rc.50 on such a machine, upgrade to this release.
This release includes the provider plugins, request processing plugins and community plugin marketplace added in rc.50. A provider plugin turns an upstream vendor into a channel type. A request processing plugin runs on requests and replies, scoped globally or to groups, channels and models. When upgrading from rc.49 or earlier, note the following.
On first start, the old sensitive word setting becomes a Request filter instance and the old setting is removed. A match now returns 400 request_rejected instead of 500 sensitive_words_detected. After rolling back to an older version, sensitive word checks are off until you fill them in again.
Multi-node deployments: manual changes to settings, permissions, plugins, channels and more reach every node within seconds. Scheduled balance updates from CHANNEL_UPDATE_FREQUENCY now run on the master node only. Finished system tasks older than 30 days are deleted automatically. To keep them longer, change the retention days in System Settings → Operations → Log Maintenance first.
From Vertex AI plugin 1.3.1, video requests go to the channel's deployment region instead of always to global. Before upgrading, check that the video models you use are available in that region.
batch_size、Imagen 的 sampleCount(不填按 4 张)和 Replicate 的 num_outputs。插件或渠道覆盖改了张数,预扣也跟着变。HOST 环境变量设置监听地址。DeepSeek 渠道的 Chat Completions 请求会把 developer 角色转成 system。修复了透传请求重试时,新渠道可能收到残缺请求体的问题。修复了切换渠道后,模型重定向同步把上一个渠道的模型名加进当前渠道的问题。修复了用输入法输入风险确认文字后按钮仍不可用的问题。修复了渠道插件设置里过长的内容把设置区撑出渠道编辑抽屉的问题。登录页的协议同意文字现在会翻译。插件运行时升级到 MoeForge moejs v0.1.0。batch_size, Imagen sampleCount (4 when omitted) and Replicate num_outputs. When a plugin or channel override changes the count, the pre-charge follows it.HOST environment variable to set the listen address. DeepSeek channels map the developer role to system in Chat Completions requests. Fixed the problem of a retried pass-through request sending an incomplete body to the next channel. Fixed the problem of model redirect sync adding the previous channel's model names after switching channels. Fixed the problem of the risk confirmation button staying disabled after typing the confirmation text with an input method. Fixed the problem of long content in a channel's plugin settings pushing the settings out of the channel drawer. The legal consent text on sign-in pages is now translated. The plugin runtime is now MoeForge moejs v0.1.0.rc.50 的变更列表 / rc.50 change list: https://github.com/QuantumNous/new-api/releases/tag/v1.0.0-rc.50
Full Changelog: v1.0.0-rc.50...v1.0.0-rc.51
Important 新增提供商插件、请求处理插件和社区插件市场。提供商插件把一个上游厂商接成一种渠道;请求处理插件在请求和回复上运行,可以按全局、分组、渠道和模型生效。
Important
新增提供商插件、请求处理插件和社区插件市场。提供商插件把一个上游厂商接成一种渠道;请求处理插件在请求和回复上运行,可以按全局、分组、渠道和模型生效。
原敏感词设置在首次启动时迁成一个「请求过滤」实例,旧设置项会被删除。命中时返回 400 request_rejected,原来是 500 sensitive_words_detected。回滚到旧版本后敏感词检查是关闭的,需要重新填写。
多节点部署:设置、权限、插件、渠道等手动改动几秒内同步到所有节点。CHANNEL_UPDATE_FREQUENCY 的定时余额更新改为只在主节点运行。超过 30 天的已结束系统任务会被自动删除,需要保留更久的请先在 系统设置 → 运维 → 日志维护 里改保留天数。
Vertex AI 插件 1.3.1 起,视频请求发到渠道填的部署地区,不再固定发到 global。升级前请确认所用的视频模型在该地区可用。
Added provider plugins, request processing plugins and a community plugin marketplace. A provider plugin turns an upstream vendor into a channel type. A request processing plugin runs on requests and replies, scoped globally or to groups, channels and models.
On first start, the old sensitive word setting becomes a Request filter instance and the old setting is removed. A match now returns 400 request_rejected instead of 500 sensitive_words_detected. After rolling back to an older version, sensitive word checks are off until you fill them in again.
Multi-node deployments: manual changes to settings, permissions, plugins, channels and more reach every node within seconds. Scheduled balance updates from CHANNEL_UPDATE_FREQUENCY now run on the master node only. Finished system tasks older than 30 days are deleted automatically. To keep them longer, change the retention days in System Settings → Operations → Log Maintenance first.
From Vertex AI plugin 1.3.1, video requests go to the channel's deployment region instead of always to global. Before upgrading, check that the video models you use are available in that region.
batch_size、Imagen 的 sampleCount(不填按 4 张)和 Replicate 的 num_outputs。插件或渠道覆盖改了张数,预扣也跟着变。HOST 环境变量设置监听地址。DeepSeek 渠道的 Chat Completions 请求会把 developer 角色转成 system。修复了透传请求重试时,新渠道可能收到残缺请求体的问题。修复了切换渠道后,模型重定向同步把上一个渠道的模型名加进当前渠道的问题。修复了用输入法输入风险确认文字后按钮仍不可用的问题。登录页的协议同意文字现在会翻译。插件运行时升级到 MoeForge moejs v0.1.0。batch_size, Imagen sampleCount (4 when omitted) and Replicate num_outputs. When a plugin or channel override changes the count, the pre-charge follows it.HOST environment variable to set the listen address. DeepSeek channels map the developer role to system in Chat Completions requests. Fixed the problem of a retried pass-through request sending an incomplete body to the next channel. Fixed the problem of model redirect sync adding the previous channel's model names after switching channels. Fixed the problem of the risk confirmation button staying disabled after typing the confirmation text with an input method. The legal consent text on sign-in pages is now translated. The plugin runtime is now MoeForge moejs v0.1.0.Full Changelog: v1.0.0-rc.49...v1.0.0-rc.50
Important 增加了 gpt-image-2.6 的支持。它的新参数 complexity、svg、frame_count、frame_duration_ms 和 markup 会转发给上游,可以生成 SVG 和 GIF,编辑图片时可以带 markup 标注图层。
Important
增加了 gpt-image-2.6 的支持。它的新参数 complexity、svg、frame_count、frame_duration_ms 和 markup 会转发给上游,可以生成 SVG 和 GIF,编辑图片时可以带 markup 标注图层。
修复了 Redis 集群和分片 Redis 上修改用户或令牌时报 keys must in same slot 的问题。多节点部署请把所有节点一起升级,新旧版本混跑时,一个节点上的用户和令牌变更不一定能及时同步到另一个版本的节点。
SQLITE_PATH 只写文件路径时,现在和默认配置一样开启 WAL 和写入等待。数据库旁边会多出 -wal 和 -shm 文件,备份时要一起复制。
模型定价保存时会逐个模型列出当前价格和新价格,确认后才保存。删掉的价格在保存前可以撤销。
Added support for gpt-image-2.6. Its new parameters complexity, svg, frame_count, frame_duration_ms, and markup are forwarded to the upstream, so it can generate SVG and GIF images, and image edits can include a markup layer.
Fixed the problem of keys must in same slot errors when editing users or tokens on Redis Cluster or sharded Redis. Upgrade every node of a multi-node deployment together. While old and new versions run side by side, a user or token change on one node may not reach nodes on the other version in time.
A SQLITE_PATH that is only a file path now gets WAL and the busy timeout, the same as the default setting. -wal and -shm files appear next to the database. Copy them together with it when you back up.
Saving Model Pricing now lists the current and new price of each model and saves only after you confirm. A deleted price can be undone until you save.
/v1 地址调用了,比如 TypeSafe 的 POST /v1/systemone。/v1/tools/search 和 /v1/tools/fetch。修复了 Vertex AI 服务账号每个请求都重新换取访问令牌的问题,令牌现在缓存到过期前 5 分钟。修复了多密钥渠道的密钥全部失败后,重新启用其中一个密钥,渠道仍然处于自动禁用的问题。修复了手动禁用的渠道在请求失败后变成自动禁用、又被自动测试重新启用的问题。渠道亲和性里的会话规则可以单独清空自己的缓存,刷新统计和清空缓存的说明也写清楚了。SQLITE_PATH 只写文件路径时丢掉 WAL 和写入等待设置,并发写入报 "database is locked" 的问题。修复了用邀请码通过 OAuth 注册的新用户没有记录邀请人的问题。/v1 paths too, such as TypeSafe's POST /v1/systemone./v1/tools/search and /v1/tools/fetch. Fixed the problem of Vertex AI service accounts requesting a new access token for every request. Tokens are now cached until 5 minutes before they expire. Fixed the problem of a multi-key channel staying auto-disabled when one key was enabled again after all keys failed. Fixed the problem of a manually disabled channel becoming auto-disabled after a failed request, which let automatic tests enable it again. Session rules in Channel Affinity can clear their own cache, and the refresh and clear actions are explained more clearly.SQLITE_PATH with only a file path dropping the WAL and busy timeout settings, which made concurrent writes fail with "database is locked". Fixed the problem of new OAuth users who signed up with an invite code not recording the inviter.Full Changelog: v1.0.0-rc.48...v1.0.0-rc.49
Important Redis 里的用户、令牌和会话缓存加了签名。多节点部署请把所有节点一起升级,新节点遇到旧版本写的缓存会回数据库重新读取。
Important
Redis 里的用户、令牌和会话缓存加了签名。多节点部署请把所有节点一起升级,新节点遇到旧版本写的缓存会回数据库重新读取。
任务插件可以用厂商自己的接口地址调用了,原来带前缀的地址照常可用。原生请求按客户端写的原样发给厂商。可灵、Vidu、MiniMax 和火山方舟的请求带 callback_url 时会被拒绝。
增加了任务模型的价格提醒。插件改动后没法计费的模型,会在模型定价里标成「价格待处理」。
User, token, and session caches in Redis are now signed. Upgrade every node of a multi-node deployment together. A new node reads from the database again when it finds a cache written by an older version.
Task plugins can now be called at the vendors' own API paths. The prefixed paths still work. Native requests go to the vendor exactly as the client wrote them. Kling, Vidu, MiniMax, and Volcengine Ark requests with callback_url are rejected.
Added price reminders for task models. Models a plugin can no longer charge are marked "Price needs attention" in Model Pricing.
/v1/videos/text2video、Vidu 的 /ent/v2/text2video,回复保持厂商的格式和错误码。即梦改用官方的 Bearer 接口。增加了可灵 4.0,时长 3 到 30 秒,按可灵返回的金额扣费。用 /v1/videos 调可灵时,尺寸按宽高比选最接近的官方比例。增加了 Vertex AI 渠道的 Gemini Omni 视频生成,按 Vertex 报告的 token 计费,需要服务账号 JSON 密钥。RELAY_HTTP2_PING_IDLE_TIMEOUT 开启,默认关闭。创建渠道的接口会返回新渠道的 ID,批量创建时每一行对应一个 ID。修复了渠道亲和的重试没有从最高优先级开始的问题。[DONE] 没被当成流结束的问题。修复了 stop 数组里混有非字符串时转成 Claude 出错的问题。修复了转成 Claude 时工具参数里的大整数被改掉的问题。修复了 Gemini 同时返回文字和工具调用时丢掉文字的问题。修复了 Bedrock Invoke 丢掉 safeguards 的问题。修复了 Cloudflare 向量回复缺少 data 的问题。修复了 xAI 流式用量丢掉 token 明细的问题。改写请求时只改需要改的字段,其余部分保持客户端原样,字段顺序和数字不再变。/v1/videos/text2video and Vidu's /ent/v2/text2video. Replies keep the vendor's format and error codes. Jimeng now uses the official Bearer API. Added Kling 4.0, with durations of 3 to 30 seconds, charged at the amount Kling returns. When Kling is called through /v1/videos, the size maps to the closest official aspect ratio. Added Gemini Omni video generation on Vertex AI channels, billed by the tokens Vertex reports. It needs a service account JSON key.RELAY_HTTP2_PING_IDLE_TIMEOUT to turn it on. Creating channels returns the new channel IDs, one for each input line in a batch. Fixed the problem of a channel affinity retry not starting from the highest priority.[DONE] line not ending the stream. Fixed the problem of a stop array with non-string items failing on the way to Claude. Fixed the problem of large integers in tool arguments changing on the way to Claude. Fixed the problem of Gemini text being dropped when it came with tool calls. Fixed the problem of Bedrock Invoke dropping safeguards. Fixed the problem of Cloudflare embedding replies missing data. Fixed the problem of xAI stream usage losing its token details. Request rewrites change only the fields they need, so the rest keeps the client's field order and numbers.Full Changelog: v1.0.0-rc.47...v1.0.0-rc.48
Important 增加了普通渠道和任务插件一起选择渠道的功能。同一个模型名,插件和普通渠道都有的时候,以前请求都发给插件,现在按优先级和权重选择渠道。
Important
增加了普通渠道和任务插件一起选择渠道的功能。同一个模型名,插件和普通渠道都有的时候,以前请求都发给插件,现在按优先级和权重选择渠道。
这个模型的定价如果还是按任务插件写的,普通渠道会报价格错误,不会扣额度。可以单独给普通渠道设置价格。
增加了 OpenAI 12 小时提示缓存单独计价的功能,按输入价格的 2.5 倍收费。
修复了视频任务超过 24 小时就被退款的问题。默认等到 72 小时才退款。
Added selecting a channel from both a regular channel and a task plugin. On image, video, and Responses endpoints, a model both can serve used to send every request to the plugin. The channel is now selected by priority and weight.
If that model's price is still written for the task plugin, a regular channel returns a price error and does not charge. Set a regular price for it in pricing.
Added a separate price for OpenAI's 12-hour prompt cache, at 2.5 times the input price. It was charged at the 30-minute rate.
Fixed the problem of video tasks being refunded after 24 hours. They now wait 72 hours. If you set TASK_TIMEOUT_MINUTES yourself, that time stays.
Full Changelog: v1.0.0-rc.46...v1.0.0-rc.47
Important 修复 CVE-2026-111943 ,请尽快升级。这一版用 Go 1.26.10。Go 校验证书时,会拒绝一小批已经不再可信的中间证书。用源码编译的,Go 也要换成 1.26.10。
Important
修复 CVE-2026-111943,请尽快升级。这一版用 Go 1.26.10。Go 校验证书时,会拒绝一小批已经不再可信的中间证书。用源码编译的,Go 也要换成 1.26.10。
Fixes CVE-2026-111943. Please upgrade soon. This release runs on Go 1.26.10. When Go checks a certificate, it rejects a small set of intermediate certificates that are no longer trusted. Builds from source need Go 1.26.10 too.
Full Changelog: v1.0.0-rc.45...v1.0.0-rc.46
Important v1 正式版即将发布,现在是最后几个 RC 版本。新增 DeepSeek V4.1 Flash、Search 和 Fetch,通知分开开关,以及拖动排列渠道优先级的功能。
Important
v1 正式版即将发布,现在是最后几个 RC 版本。新增 DeepSeek V4.1 Flash、Search 和 Fetch,通知分开开关,以及拖动排列渠道优先级的功能。
敏感词过滤默认关着。以前填过敏感词、但没保存过这个开关的,升级后到「请求检查」里打开。
v1 is coming soon. These are the last few release candidates. Added DeepSeek V4.1 Flash, Search and Fetch, separate notification switches, and dragging to set channel priority.
Sensitive word filtering is off by default. If you filled in a word list and never saved the switch, turn it on under Request checks after upgrading.
@thinking 和 @effort 也能用。新增 Search 和 Fetch,地址是 /v1/tools/search 和 /v1/tools/fetch。Claude Code 和 Codex 让 DeepSeek 去搜索、打开网页时,按次数计费,每 1000 次 0.75 美元。@thinking and @effort work too. Added Search and Fetch at /v1/tools/search and /v1/tools/fetch. When Claude Code or Codex has DeepSeek search or open a page, that is charged per request, $0.75 per 1,000.Full Changelog: v1.0.0-rc.44...v1.0.0-rc.45
Important v1 Coming Soon & Log Export / v1 即将发布与日志导出
Important
v1 Coming Soon & Log Export / v1 即将发布与日志导出
v1 正式版即将发布,现在是最后几个 RC 版本。
使用日志支持导出 CSV。可以在日志维护里设置是否允许用户导出,以及日志最长时间范围,默认 365 天。普通用户的日志查询、导出和数据看板都按这个范围限制。
游乐场的数据按账号分开保存,升级后之前的聊天记录会清空一次。
Claude 请求按模型支持的档位发送思考强度。发给 Gemini 的工具和结构化输出保留完整的 JSON Schema。
v1 is coming soon. These are the last few release candidates.
Usage logs can be exported as CSV. In Log Maintenance you can choose whether users may export and set the longest log time range, 365 days by default. Regular users' log searches, exports, and dashboard charts all follow that range.
Playground data is stored per account. Earlier Playground history is cleared once after the upgrade.
Claude requests get the thinking effort each model accepts. Tools and structured output sent to Gemini keep their full JSON Schema.
max or xhigh, on Sonnet 4.5 and Haiku 4.5, and when a model suffix or modifier turns thinking off. Native Claude requests without a suffix or modifier pass through as sent.temperature, top_p, and top_k for it. Gemini 2.5 Pro gets a 32768 thinking budget for xhigh and max.const, oneOf, $ref, $defs, type arrays, and non-string enums. A schema that Gemini's typed Schema can hold is sent as before, and any other schema goes in parametersJsonSchema or responseJsonSchema. Models set to Gemini API v1 get a cleaned schema and a conversion warning (#7455, #7704).image_url, so vLLM and strict gateways accept Codex screenshots and tool images (#7578).thinking.block_binding and every anthropic-beta header value. This fixes "The block is bound to a different conversation" in multi-turn conversations whose system prompt or tools change (#7732).Full Changelog: v1.0.0-rc.43...v1.0.0-rc.44
Important 完善了 i18n,控制台、API 错误、邮件、通知和服务器日志都支持多语言。使用 DEFAULT_LANGUAGE 环境变量设置默认语言,不设置时为英文。
Important
完善了 i18n,控制台、API 错误、邮件、通知和服务器日志都支持多语言。使用 DEFAULT_LANGUAGE 环境变量设置默认语言,不设置时为英文。
渠道新增“忽略上游证书校验”开关,可以连接使用自签证书的上游。
任务插件新增 utils.fetch,可以读取上游数据。
豆包 Seedance 1.5 Pro、2.x 和万相 3.0 支持传 -1 自动选择时长,按上游回报的时长计费。
Improved i18n. Console messages, API errors, emails, notifications, and server logs are now multilingual. Set the default language with the DEFAULT_LANGUAGE environment variable. It defaults to English.
Channels have a new Skip Upstream TLS Verification switch for upstreams that use self-signed certificates.
Task plugins can read upstream data with the new utils.fetch.
Doubao Seedance 1.5 Pro and 2.x, and Wan 3.0, accept -1 to pick the video duration automatically and bill the duration the upstream reports.
DEFAULT_LANGUAGE (en, zh-CN, or zh-TW) sets the default language of backend messages and server logs. It defaults to en.utils.fetch in the submit, query, and content-download hooks. It reaches the channel Base URL and the additional request domains the plugin declares, and the upload dialog lists those domains. The Task Plugin API v1 document lists the hooks and limits.utils.json.clone and export parseSubmitEventDelta with no capability declaration. Plugins that declare json-clone@1 or submit-sse-delta@1 keep loading.-1 for Seedance 1.5 Pro and 2.x and for Wan 3.0. The request reserves the model's longest duration (12 seconds for Seedance 1.5 Pro, 15 for 2.0, 30 for 2.5 and Wan 3.0) and bills the duration the upstream reports. A JSON duration on /v1/videos now reaches Ark, and Wan 3.0 smart duration works on /v1/video/generations. (#7700)trailing_assistant_omitted warning. Codex sends such input when it retries a stream that failed after a reply finished. Claude 4.6 and later, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite and later rejected it with a 400. Chat Completions and Claude Messages requests keep their prefill.Full Changelog: v1.0.0-rc.42...v1.0.0-rc.43
Important Search Billing & Task Plugins / 搜索计费与任务插件
Important
Search Billing & Task Plugins / 搜索计费与任务插件
重做了搜索函数计费。渠道可以按上游自己的方式进行网络搜索,并按上游回报的搜索次数计费。
任务插件性能大幅提升。支持插件决定是否保留 JSON 字段顺序,声明 json-order@1 即可;其他插件行为不变。
高级自定义渠道可以把 Claude Messages 和 Gemini 转到 OpenAI Responses。
新增 xAI Grok Imagine 视频任务插件。
Search-function billing has been reworked. Channels can encode Web search for their upstream and settle from the search counts the vendor reports.
Task Plugins are substantially faster. A plugin that must keep the client’s JSON field order can declare json-order@1. Other plugins are unchanged.
Advanced Custom channels can convert Claude Messages and Gemini to OpenAI Responses. An xAI Grok Imagine video Task Plugin is added.
/v1/responses unchanged. Kimi currently serves kimi-k3 on that path.grok-imagine-video, grok-imagine-video-1.5, and grok-imagine-video-1.5-lite.json-order@1 keep the client’s JSON field order through to the upstream.exec and apply_patch are kept on Claude and Gemini upstreams (#7636).-high unchanged.gpt-6-sol and gpt-6-luna use max_completion_tokens (#7559).Full Changelog: v1.0.0-rc.41...v1.0.0-rc.42
Important Access Tokens & Task Plugins / 访问令牌与任务插件
Important
Access Tokens & Task Plugins / 访问令牌与任务插件
安全与访问里的单个面板访问令牌,改为最多 20 个具名访问令牌。创建时只显示一次,默认 30 天过期;作用域按本人已有权限勾选。升级后旧版令牌还能再用 30 天,之后会被拒绝,且不能重新生成,请在到期前换成新令牌。
管理员管理用户时,删除、改状态或角色、设置密码或权限、创建管理员、重置通行密钥、关闭两步验证、清除绑定,都需要一次性校验。额度调整,以及修改显示名、分组、备注,不需要。
任务插件改由 moejs 运行,性能大幅提升,插件接口不变。
豆包 Seedream 图片改走任务插件的 OpenAI Images(/v1/images/generations、/v1/images/edits)。新增 Seedream 5.0 Flash;各模型按自己的档位计价。任务插件收窄用量项之后,已保存的价格仍可编辑。
Security & Access replaces the single dashboard Access Token with up to 20 named Access tokens. A token is shown once at creation, expires in 30 days by default, and its Scopes stay within what the owner can already do. The Legacy token keeps working for 30 days after the upgrade, then it is rejected and cannot be regenerated. Replace it before then.
Admin user management now requires a single-use verification for delete, status or role changes, password or Permissions, creating an admin, resetting a Passkey, disabling 2FA, and clearing a binding. Quota adjustments, and edits to display name, Group, or remark, do not.
Task Plugins now run on moejs, with a large performance improvement.
Doubao Seedream images now go through the Task Plugin OpenAI Images paths (/v1/images/generations, /v1/images/edits). Seedream 5.0 Flash is added, and each model is priced on its own tiers. Prices saved before a Task Plugin narrows its usage facts stay editable.
exec and apply_patch are no longer dropped.output_config (#7598, #7561).v0.0.0 (#7536).Full Changelog: v1.0.0-rc.39...v1.0.0-rc.41
任务插件源码上限提高到 8 MiB,更大的插件可以正常上传和保存。 上游提交返回 201 / 202 等成功状态时,任务会正常继续,不再被当成失败。
Important
Task Plugins / 任务插件
任务插件源码上限提高到 8 MiB,更大的插件可以正常上传和保存。
上游提交返回 201 / 202 等成功状态时,任务会正常继续,不再被当成失败。
定价配置在任务插件收窄可选项后,仍可编辑剩下的档位和价格,不会整表打不开。
Claude / Responses 里工具结果中的图片会按图片送给上游,不再当成文本。
Task Plugin source uploads now allow up to 8 MiB, so larger plugins can be saved.
A 2xx upstream submit status such as 201 / 202 is accepted, so the task continues instead of failing.
Pricing Configuration stays editable after a Task Plugin narrows its options; remaining tiers keep their prices.
Images in Claude / Responses tool results are sent upstream as images, not as text.
tool_result images and Responses tool-output media upstream as images, not as text (#7512, #7510).Full Changelog: v1.0.0-rc.39...v1.0.0-rc.40
Important Task Plugins & Quota Settings / 任务插件与额度设置
Important
Task Plugins & Quota Settings / 任务插件与额度设置
任务插件现已兼容 OpenAI Images:
可承接 /v1/images/generations 和 /v1/images/edits。
请求内提交任务并等待完成后,按 OpenAI 图片格式返回。
阿里云百炼万相图片模型改走任务插件;豆包 Seedream 图片生成已接入。
New API 渠道可绑定多个上游任务插件,用于对接另一套网关里的多个任务插件。
渠道的模型重定向更清晰,应用上游模型变更前可先预览。
额度设置可配置钱包免预扣门槛和输入预扣倍率,余额足够时跳过预扣。
Task Plugins now serve OpenAI Images:
/v1/images/generations and /v1/images/edits.
The host submits the task, waits inside the request, and returns an OpenAI image response.
Unclaimed models still use the existing image relay.
Alibaba Bailian Wanxiang image models now go through Task Plugins.
Doubao Seedream image generation is added.
A New API Channel can bind multiple Upstream task plugins, for another gateway that hosts several Task Plugins.
Model Mapping is clearer, and upstream model changes can be previewed before apply.
Quota Settings add Wallet pre-consume bypass threshold, and Input pre-consume multiplier.
High enough balances skip pre-consume.
/v1/images/generations, /v1/images/edits).reasoning_effort to Zhipu (#7124).Full Changelog: v1.0.0-rc.38...v1.0.0-rc.39
Important Request Policies & Channels / 请求策略与渠道
Important
Request Policies & Channels / 请求策略与渠道
系统设置新增请求策略:集中配置请求检查、会话与重试、渠道健康,并在使用日志中留下策略决策记录。
渠道创建与编辑补充快捷选项、透传说明,以及模型重定向工作台。高级自定义渠道可按路径单独开启透传。新增 vLLM、SGLang 渠道;Ollama 渠道可按渠道开启 OpenAI 兼容对话。
渠道可开启 Responses WebSocket。高级自定义、Sub2API、New API 渠道在开关打开后可用;高级自定义仅用于无转换的 /v1/responses。
System Settings now includes Request policies: Request checks, Sessions and retries, and Channel health, with policy decision records on Usage Logs.
Channel create and edit add Quick options, Pass-through guidance, and a Model Mapping workbench. Advanced Custom channels can enable Pass-through per path. vLLM and SGLang channels are added. Ollama channels can enable OpenAI-compatible chat per Channel.
Channels can enable Responses WebSocket. Advanced Custom, Sub2API, and New API channels can use it when the toggle is on. Advanced Custom only uses converter-free /v1/responses paths.
Warning
Experimental Plugin System / 实验性插件系统
自 v1.0.0-rc.27 起引入的实验性插件系统仍处于测试阶段,并涉及较大的架构调整。如遇问题,请及时反馈。
从 v1.0.0-rc.26 或更早版本升级的用户,必须重新配置所有视频模型价格;已完成该配置的用户无需重复操作。本版本仍不推荐用于生产环境。
The experimental plugin system introduced in v1.0.0-rc.27 remains under testing and includes major architectural changes. Please report any issues promptly.
Users upgrading from v1.0.0-rc.26 or earlier must reconfigure all video model prices. Users who have already completed this step do not need to repeat it. This release is still not recommended for production use.
/v1/responses.:countTokens as an unknown path instead of treating it as generateContent (#7388).Full Changelog: v1.0.0-rc.37...v1.0.0-rc.38
Legacy pricing will be deprecated soon. Pricing Configuration can convert current prices to an Expression draft; review the effective prices, then sav…
Important
Pricing Configuration & Task Plugins / 定价配置与任务插件
旧定价模式即将弃用。定价配置可将当前价格转换为计费表达式草稿,预览生效价格后再保存。新编辑器支持按条件计价、按次、含时间条件,以及图片用量与图片缓存。gpt-image-2、gpt-image-2.5-sunburst、gpt-image-2.5-flare 已有内置计费表达式默认值;已有管理员覆盖仍然优先。
任务插件能力增强:支持流式输出;定价配置可按任务插件设置用量价格与计费表达式;插件详情展示更新日志,缺省回退英文。创建与编辑渠道时,任务插件可参与提供商与模型选择,并带出对应插件信息。
渠道的创建与编辑已统一:共用提供商选择、分页配置、模型发现、模型重定向与参数覆盖。创建渠道时会优先使用已可用的任务插件。
安全与访问支持多个通行密钥域名。通行密钥只认创建它的网站。移除仍在使用的域名会先预览受影响数量,并要求确认;被移除域名上的通行密钥将无法继续使用。
Legacy pricing will be deprecated soon. Pricing Configuration can convert current prices to an Expression draft; review the effective prices, then save to apply. The editors now support Pricing by condition, Per-request, Time-based, and image usage / image cache. gpt-image-2, gpt-image-2.5-sunburst, and gpt-image-2.5-flare have built-in Expression defaults; existing administrator overrides still take precedence.
Task Plugins capabilities are expanded: Streaming; Usage prices and Expression in Pricing Configuration per Task Plugin; and Changelog in plugin details with an English fallback. When creating or editing a Channel, Task Plugins can take part in provider and model selection and bring along the matching plugin info.
Channel create and edit are unified: shared provider selection, tabbed setup, model discovery, Model Mapping, and parameter overrides. Channel creation prefers an available Task Plugin.
Security & Access supports multiple Passkey domains. Passkeys only work on the website they were created for. Removing a domain still in use previews how many Passkeys are affected and requires confirmation; Passkeys on a removed domain will stop working.
Warning
Experimental Plugin System / 实验性插件系统
自 v1.0.0-rc.27 起引入的实验性插件系统仍处于测试阶段,并涉及较大的架构调整。如遇问题,请及时反馈。
从 v1.0.0-rc.26 或更早版本升级的用户,必须重新配置所有视频模型价格;已完成该配置的用户无需重复操作。本版本仍不推荐用于生产环境。
The experimental plugin system introduced in v1.0.0-rc.27 remains under testing and includes major architectural changes. Please report any issues promptly.
Users upgrading from v1.0.0-rc.26 or earlier must reconfigure all video model prices. Users who have already completed this step do not need to repeat it. This release is still not recommended for production use.
gpt-image-2, gpt-image-2.5-sunburst, and gpt-image-2.5-flare, including separate text and image cache prices. Existing administrator overrides stay in place./api/status requests (#7189).Full Changelog: v1.0.0-rc.36...v1.0.0-rc.37
v1.0.0-rc.37- Pricing Configuration, Task Plugins, and Passkey Latest
Latest
Compare
Important Task Plugins & Pricing / 任务插件与定价
Important
Task Plugins & Pricing / 任务插件与定价
本版本完善任务插件与插件市场:插件图标、插件官网、插件元数据与能力,创建渠道时可带出对应插件信息。同时改进模型管理、目录展示与广场展示、定价配置,以及 API 密钥与用户额度、使用日志、兑换码。定价货币可切换为站点货币以便录入和查看,额度仍按美元扣费。
This release improves Task Plugins and the Marketplace: Plugin icon, Plugin website, Plugin metadata, and Capabilities, plus matching plugin info when creating a Channel. It also improves Model management, Catalog visibility, Model square visibility, Pricing Configuration, Quota on API Keys and users, Usage Logs, and Redemption Codes. Pricing currency can switch to Site currency for entry and review; Quota is still deducted in USD.
Warning
Experimental Plugin System / 实验性插件系统
自 v1.0.0-rc.27 起引入的实验性插件系统仍处于测试阶段,并涉及较大的架构调整。如遇问题,请及时反馈。
从 v1.0.0-rc.26 或更早版本升级的用户,必须重新配置所有视频模型价格;已完成该配置的用户无需重复操作。本版本仍不推荐用于生产环境。
The experimental plugin system introduced in v1.0.0-rc.27 remains under testing and includes major architectural changes. Please report any issues promptly.
Users upgrading from v1.0.0-rc.26 or earlier must reconfigure all video model prices. Users who have already completed this step do not need to repeat it. This release is still not recommended for production use.
icon.svg / icon.png), integrity checks, Capabilities, and clearer install/update flows.Full Changelog: v1.0.0-rc.35...v1.0.0-rc.36
v1.0.0-rc.36 - Task Plugins, Pricing Configuration, and Quota
Compare
Important Task Plugin Controls / 任务插件控制简化
Important
Task Plugin Controls / 任务插件控制简化
任务插件现在只有两层开关:总开关与每个插件自己的开关。关闭某个插件后,不会再回退到同名内置插件,同名模型也不会再继续由内置插件处理。内置插件会按请求中的模型名回显(遵循模型重定向),而不是上游真实 ID。
Task plugins now have two control layers only: the master switch and each plugin’s own toggle. Disabling a plugin no longer falls back to a same-name built-in plugin, so same-name models are no longer still handled by the built-in plugin. Built-in plugins echo the requested model name (honoring model redirects) instead of the upstream model ID.
Warning
Experimental Plugin System / 实验性插件系统
自 v1.0.0-rc.27 起引入的实验性插件系统仍处于测试阶段,并涉及较大的架构调整。如遇问题,请及时反馈。
从 v1.0.0-rc.26 或更早版本升级的用户,必须重新配置所有视频模型价格;已完成该配置的用户无需重复操作。本版本仍不推荐用于生产环境。
The experimental plugin system introduced in v1.0.0-rc.27 remains under testing and includes major architectural changes. Please report any issues promptly.
Users upgrading from v1.0.0-rc.26 or earlier must reconfigure all video model prices. Users who have already completed this step do not need to repeat it. This release is still not recommended for production use.
wan3.0-video and wan3.0-video-prime all-in-one video models, with DashScope media, resolution/ratio, duration rules, and billing ratios.duration / seconds = -1): submit -1 upstream, pre-consume 30 seconds, then settle from actual usage. Other Wan models reject this marker instead of silently defaulting to 5 seconds.kimi-k3 to the Moonshot model list.tool_choice=required no longer fails for missing tools (#7235).xl breakpoint.common and relaykit JSON helpers through a host-injectable codec so the JSON engine can be swapped in one place.Full Changelog: v1.0.0-rc.34...v1.0.0-rc.35
v1.0.0-rc.35 — Wan 3.0 Video, Plugin Routing
Compare
One column per month.