
OpenAI 全系列模型已升级至 GPT-4 架构。LobeHub v1.6 跟进这一变化,正式支持 GPT-4o mini。对于 LobeHub Cloud 用户,这次升级更进一步:GPT-4o mini 现已成为默认模型,替代 GPT-3.5-turbo。
这意味着从第一条消息开始,你就能获得更强的对话能力,无需任何配置调整。
GPT-4o mini:能力与成本的平衡之选
GPT-4o mini 以更小的规模提供 GPT-4 级别的智能。响应速度足以支撑实时交互,能力覆盖大多数日常任务 —— 起草文案、分析数据、编程辅助和创意工作。
GPT-4o mini 的适用场景:
- 比 GPT-3.5 更强的推理能力,同时避免完整 GPT-4o 的延迟
- 高频对话的成本优化默认选项
- 指令遵循和工具调用方面的稳定表现
面对复杂推理任务时,可切换至完整 GPT-4o 或其他提供商(Claude 3.5 Sonnet、Gemini 1.5 Pro)获取最强能力。
Cloud 服务:默认模型自动升级
LobeHub Cloud 用户将自动获得此次服务升级。新对话默认以 GPT-4o mini 启动。现有用户无需更改任何设置 —— 模型切换器会自动将新默认选项置顶显示。
Cloud 现已支持:
- GPT-4o mini(默认)
- GPT-4o
- Claude 3.5 Sonnet
- Gemini 1.5 Pro
体验优化与修复
- 新增 GPT-4o mini 模型配置和参数默认值
- 更新 LobeHub Cloud 默认模型选择逻辑
- 优化模型切换器 UI,突出推荐选项
- 修复新版 OpenAI 模型流式响应的边界情况
版本详情
Features
What's improved
- v1.5.5Added Gemma2 instead of outdated Gemma.
- v1.5.1Improve brand url.
- v1.4.2Remove code related to gemini-pro-vision.
- v1.4.1Add cloud promotion banner.
- v1.3.6Improve agent runtime code.
- v1.3.4Support disable clientFetch by default.
- v1.2.14Provider changes with model in model settings.
- v1.2.11Update deepseek max token.
- v1.2.9Fix tool message suspense loading.
- v1.2.7Improve delete assistant message with tools.
- v1.2.4Update ProviderAvatar for Baichuan & Stepfun.
- v1.2.3Refactor the portal implement.
- v1.2.1Refactor the portal file.
- v1.1.18Add Gemini 1.5 stable version model & 2M context window.
- v1.1.17Refactor to prepare for Chat Portal.
- v1.1.14Refactor plugin state.
- v1.1.11Refactor format utils.
- v1.1.9Always show action on mobile.
- v1.1.7Add AES-GCM footer and tooltip.
- v1.1.6Refactor the server db implement.
- v1.1.5Remove deprecated env.
- v1.1.1Fixed System Agent missing in mobile layout.
- v1.0.11Fix phi3 icon display under OpenRouter.
- v1.0.8Update brand and improve docs.
- v1.0.7Remove some i18n text.
- v1.0.6Fix footer being cut on wide screen.
- v1.0.4Add stepfun as a new provider.
- v1.0.1Remove r2 cdn url.
What's fixed
- v1.5.4Fix delete session group.
- v1.5.3Fix OpenAI deployment restrictions, fix cant duplicate assistant.
- v1.5.2Fix session not reorder after send message.
- v1.4.3Save assistant info on blur.
- v1.3.5Fix assistant meta change race issue.
- v1.3.3Allow user to use their own WebRTC signaling.
- v1.3.2Automatic refresh when change language.
- v1.2.13Fix tool message order.
- v1.2.12Fixed mobile web page navigation issue with inbox assistant, support to disable clerk signup with feature flag.
- v1.2.10Improve tools calling error feedback when arguments are not correctly.
- v1.2.8Allow builtin tools to trigger AI message.
- v1.2.6Clerk provider refreshes continously.
- v1.2.5Fix clerk appearance is not applied correctly.
- v1.2.2Display issue when select default model in System Agent.
- v1.1.16Fix clerk UNAUTHORIZED error after long-time hang-up.
- v1.1.15Pin @azure/core-rest-pipeline@1.16.0 to fix azure openai issue.
- v1.1.13-check_updates cannot be set by FEATURE_FLAGS.
- v1.1.12Fix azure tools calling.
- v1.1.9Fix agent tags.
- v1.1.8Fix anthropic parallel tools calling.
- v1.1.6Fix incorrect baseURL for Groq in client mode.
- v1.1.4Create first-time user on server db.
- v1.1.3Ollama not enabled client fetch by default.
- v1.0.14Fix to send image without text.
- v1.0.13Fix and improve tool calling.
- v1.0.12Fix auto avatar.
- v1.0.10Typo in reset settings.
- v1.0.8Fix hydration mismatch on macOS.
- v1.0.5Fix onboard and auth for community version.
- v1.0.3Fix clerk UNAUTHORIZED auth error.
- v1.0.2Openai key and openai proxy are invalid in feature flags.
