{"success":true,"message":"Success","data":{"items":[{"announcementId":"ann_aiGzRIQXU2u1hXQ-","titleZh":"首充翻倍 · 充 $10 到账 $20，充 $100 到账 $200.  查看活动规则","titleEn":"Double Your First Top-Up · Pay $10, get $20. Pay $100, get $200. View Terms","summaryZh":"即日起，首个充值订单可获等额赠送额度：充 10送10送，充 100送100，赠送上限 $100。赠送额度自入账起 30 天内有效，按模型折扣价扣费，并优先于充值余额使用。","summaryEn":"Starting today, your first top-up order gets an equal bonus credit: top up 10 and get 10 free, top up 100 and get 100 free, with a bonus cap of $100. The bonus credit is valid for 30 days from the date it's credited, is charged at the model's discounted rate, and is used before your topped-up balance.","contentZh":"<p>从今天起，HiLinkup 新账户的首次充值会获得与充值金额 1:1 等额的赠送额度。充多少，账户里就多出同样一笔可用的额度——充 10到账20，充 100到账200。</p><p></p><p>赠送额度不是体验金，它与充值余额一样可以调用平台全部 200+ 模型，也同样适用模型折扣价。</p><p></p><h3><strong>活动规则</strong></h3><ol><li><p><strong>适用对象</strong>：所有 HiLinkup 账户，每个账户仅限首次充值享受一次；后续充值不再赠送。</p></li></ol><ol><li><p><strong>赠送比例</strong>：1:1 等额赠送。</p></li></ol><ol><li><p><strong>赠送上限</strong>：单账户最高赠送 100。单次充值超过100。单次充值超过100 的部分不再产生赠送。</p></li></ol><ol><li><p><strong>有效期</strong>：赠送额度自入账之日起 30 天内有效，到期未使用部分自动失效，不予折现或延期。</p></li></ol><ol><li><p><strong>扣费顺序</strong>：账户内同时存在赠送额度与充值余额时，默认优先扣除赠送额度，以便你在有效期内先用掉即将过期的部分。</p></li></ol><ol><li><p><strong>折扣同享</strong>：赠送额度按模型的折扣后价格计费，不按原价折算。例如<strong>DeepSeek V4.1 Flash 1017</strong>模型折扣价为 4.8 折，使用赠送额度调用该模型时同样按 4.8 折扣费。</p></li></ol><p></p><h3><strong>常见问题</strong></h3><p><strong>赠送额度用不完会怎样？</strong><br>有效期 30 天，到期未使用部分自动失效。建议优先用赠送额度跑高频、低成本的调用。</p><p><strong>分两次充值，赠送能叠加吗？</strong><br>不能。赠送仅对首次充值生效一次，上限 $100，与充值次数无关。</p><p><strong>赠送额度可以提现或转给其他账户吗？</strong><br>不可以。赠送额度仅限本账户调用模型使用，不可提现、不可转让、不可开具发票。</p><p><strong>充值余额有有效期吗？</strong><br>没有。充值余额长期有效，只有赠送部分受 30 天限制。</p><h3><strong>其他说明</strong></h3><p>本活动自发布之日起长期有效，如规则调整将提前在官网公告。</p><p></p>","contentMode":"RICH_TEXT","contentEn":"<p>Starting today, first-time top-ups on new HiLinkup accounts get a bonus credit equal to the top-up amount at a 1:1 ratio. Whatever you top up, your account gets the same amount in usable credit—top up 10 and you get 20, top up 100 and you get 200.</p><p></p><p>The bonus credit isn't a trial credit. Like your topped-up balance, it can be used to call all 200+ models on the platform, and it's also subject to the models' discounted prices.</p><p></p><p>Activity rules</p><p>Eligibility: All HiLinkup accounts; each account can enjoy this only once on its first top-up. Subsequent top-ups don't get the bonus.</p><p>Bonus ratio: 1:1 equal bonus.</p><p>Bonus cap: Up to 100 per account. Any amount above 100 in a single top-up doesn't generate bonus credit.</p><p>Validity: The bonus credit is valid for 30 days from the date it's credited. Any unused portion expires automatically at the end of that period and can't be cashed out or extended.</p><p>Deduction order: When an account has both bonus credit and a topped-up balance, the bonus credit is deducted first by default, so you can use up the portion that's about to expire within the validity period.</p><p>Same discounts: Bonus credit is billed at the model's discounted price, not converted at the original price. For example, the DeepSeek V4.1 Flash 1017 model has a 4.8 discount (i.e., 48% of the original price); when you use bonus credit to call that model, you're also billed at the 4.8 discount.</p><p></p><p>FAQ</p><p>What happens if I don't use up the bonus credit?<br>It's valid for 30 days, and any unused portion expires automatically. We recommend using the bonus credit first for high-frequency, low-cost calls.</p><p>If I top up twice, can the bonuses stack?<br>No. The bonus applies only once to the first top-up, capped at $100, regardless of how many times you top up.</p><p>Can the bonus credit be withdrawn or transferred to another account?<br>No. The bonus credit can only be used to call models on this account; it can't be withdrawn, transferred, or invoiced.</p><p>Does the topped-up balance have an expiration date?<br>No. The topped-up balance is valid long-term; only the bonus portion is subject to the 30-day limit.</p><p>Other notes</p><p>This activity is valid long-term from the date of publication. If the rules change, we'll announce it on the official website in advance.</p>","category":"INFO","linkUrl":"http://bonus.hilinkup.com/","isPinned":true,"popup":false,"showBanner":true,"publishedAt":"2026-10-30T15:46:13.620004","id":"ann_aiGzRIQXU2u1hXQ-","title":"首充翻倍 · 充 $10 到账 $20，充 $100 到账 $200.  查看活动规则","summary":"即日起，首个充值订单可获等额赠送额度：充 10送10送，充 100送100，赠送上限 $100。赠送额度自入账起 30 天内有效，按模型折扣价扣费，并优先于充值余额使用。","content":"<p>从今天起，HiLinkup 新账户的首次充值会获得与充值金额 1:1 等额的赠送额度。充多少，账户里就多出同样一笔可用的额度——充 10到账20，充 100到账200。</p><p></p><p>赠送额度不是体验金，它与充值余额一样可以调用平台全部 200+ 模型，也同样适用模型折扣价。</p><p></p><h3><strong>活动规则</strong></h3><ol><li><p><strong>适用对象</strong>：所有 HiLinkup 账户，每个账户仅限首次充值享受一次；后续充值不再赠送。</p></li></ol><ol><li><p><strong>赠送比例</strong>：1:1 等额赠送。</p></li></ol><ol><li><p><strong>赠送上限</strong>：单账户最高赠送 100。单次充值超过100。单次充值超过100 的部分不再产生赠送。</p></li></ol><ol><li><p><strong>有效期</strong>：赠送额度自入账之日起 30 天内有效，到期未使用部分自动失效，不予折现或延期。</p></li></ol><ol><li><p><strong>扣费顺序</strong>：账户内同时存在赠送额度与充值余额时，默认优先扣除赠送额度，以便你在有效期内先用掉即将过期的部分。</p></li></ol><ol><li><p><strong>折扣同享</strong>：赠送额度按模型的折扣后价格计费，不按原价折算。例如<strong>DeepSeek V4.1 Flash 1017</strong>模型折扣价为 4.8 折，使用赠送额度调用该模型时同样按 4.8 折扣费。</p></li></ol><p></p><h3><strong>常见问题</strong></h3><p><strong>赠送额度用不完会怎样？</strong><br>有效期 30 天，到期未使用部分自动失效。建议优先用赠送额度跑高频、低成本的调用。</p><p><strong>分两次充值，赠送能叠加吗？</strong><br>不能。赠送仅对首次充值生效一次，上限 $100，与充值次数无关。</p><p><strong>赠送额度可以提现或转给其他账户吗？</strong><br>不可以。赠送额度仅限本账户调用模型使用，不可提现、不可转让、不可开具发票。</p><p><strong>充值余额有有效期吗？</strong><br>没有。充值余额长期有效，只有赠送部分受 30 天限制。</p><h3><strong>其他说明</strong></h3><p>本活动自发布之日起长期有效，如规则调整将提前在官网公告。</p><p></p>"},{"announcementId":"ann_mPJs8SKMsbXs-Zrw","titleZh":"DeepSeek V4.1 Flash 1017 已上线，6.5折优惠","titleEn":"DeepSeek V4.1 Flash 1017 Is Live — 35% Off Official Pricing","contentZh":"<p>HiLinkup 已接入 DeepSeek V4.1 Flash 1017，现以官方价 6.5 折开放调用。<br><br>这不是一个新模型，是同一个月前发布的 V4.1 Flash 重新训练了一版权重。架构、参数量、上下文长度都没变，变的是后训练——跑分上去了，任务耗时和输出长度下来了，单个任务的实际花费随之下降。<br><br>【模型信息】<br>模型名称：DeepSeek V4.1 Flash 1017<br>模型 ID：doubao/deepseek-v4.1-flash-1017<br>上下文长度：1M tokens<br>最大输出：384K tokens<br>输入类型：文本 + 图像（原生多模态）<br>思考模式：支持，可调 reasoning effort<br><br>【这次升级改了什么】<br>1. 端到端项目生成能力增强<br>基于自反馈的强化学习训练，重点强化了从零构建完整项目的能力。在极少提示词、极少交互轮次下即可产出可运行的项目。官方演示案例包括故宫 3D 建模、网页版赛车游戏、Godot 引擎复刻《传送门 2》。<br><br>2. 输出更短，跑得更快<br>官方称之为性能与成本的帕累托改进。单价没有调整，但因为生成更少的 token，单任务账单明显下降。官方性能/成本图上，DeepSWE 单任务成本从约 0.50 美元降到约 0.20 美元。<br><br>3. 针对 Agent 平台定向优化<br>DeepSeek Harness、腾讯 CodeBuddy、WorkBuddy 均做了专项适配。<br><br>【能力提升（DeepSeek 官方数据）】<br>Terminal-Bench 4.0 31.2 → 42.3 +35.6%<br>ProgramBench 20.3 → 29.1 +43.3%<br>HLE（无工具） 36.8 → 42.6 +15.8%<br>Apex Bench v1.1 64.6 → 72.1 +11.6%<br>Agents' Last Exam 31.8 → 35.4 +11.3%<br>Automation-Bench v1.0.6 54.8 → 60.5 +10.4%<br>DeepSWE v1.1 74.2 → 77.0 +3.8%<br><br>涨幅集中在终端操作与多步 Agent 任务，静态知识型基准提升相对有限。DeepSWE v1.1 的 77.0 高于 Claude Opus 5 的 74.0 与 GPT-5.6 Sol 的 73.0。<br><br>【适合哪些场景】<br>Agent 与终端自动化：CLI 操作、依赖安装、shell 脚本、git 工作流、日志解析这类长链路任务，是本次提升最明显的方向。<br>从零搭原型：一句话描述需求，直接产出可跑的游戏、3D 场景、可视化页面。适合快速验证想法，而不是精雕细琢。<br>仓库级代码修改：多文件重构、跨模块改造、仓库结构生成。配合 1M 上下文可以直接吃下整个中型代码库。<br>截图与设计稿转代码：原生多模态，图像输入不需要额外拼接视觉模型。<br>高频批量调用：6.5 折的价格叠加更短的输出，适合跑量大、单次质量要求中等的任务。<br><br>【价格】<br>HiLinkup 定价为 DeepSeek 官方价的 6.5 折，按每 100 万 tokens 计。<br>输入（缓存未命中） 官方 0.15 美元 → HiLinkup 0.0975 美元<br>输入（缓存命中） 官方 0.003 美元 → HiLinkup 0.00195 美元<br>输出 官方 0.60 美元 → HiLinkup 0.39 美元<br><br>官方采用峰谷定价，峰时价为闲时价的 2 倍（0.30 / 0.006 / 1.20 美元）。HiLinkup 同样按 6.5 折折算，上表为闲时口径。<br><br>【如何调用】<br>OpenAI 兼容协议，只改 model 名，其余代码不用动。</p>","contentMode":"RICH_TEXT","contentEn":"<p>HiLinkup now serves DeepSeek V4.1 Flash 1017 at 35% off DeepSeek's official rates.<br><br>This is not a new model. It is the same V4.1 Flash from last month with retrained weights. Architecture, parameter count, and context length are unchanged. What changed is post-training — benchmarks went up, task runtime and output length came down, and the real cost per task dropped with them.<br><br>[MODEL INFO]<br>Model ID: doubao/deepseek-v4.1-flash-1017<br>Context length: 1M tokens<br>Max output: 384K tokens<br>Input: text + images (native multimodal)<br><br>[WHAT CHANGED]<br>1. Stronger end-to-end project generation<br>Trained with reinforcement learning on self-feedback, with a focus on building complete projects from scratch. Runs with very few prompts and very few interaction turns produce working output. DeepSeek's own demos include a 3D model of the Forbidden City, a browser-based racing game, and a Portal 2 recreation built in Godot.<br><br>2. Shorter outputs, faster runs<br>DeepSeek calls it a Pareto improvement in performance and cost. Per-token pricing did not move, but the model writes fewer tokens, so the bill per task drops noticeably. On DeepSeek's performance/cost chart, DeepSWE cost per task fell from roughly $0.50 to roughly $0.20.<br><br>[BENCHMARKS — DeepSeek's own figures]<br>Terminal-Bench 4.0        31.2 → 42.3     +35.6%<br>ProgramBench              20.3 → 29.1     +43.3%<br>HLE (no tools)            36.8 → 42.6     +15.8%<br>Apex Bench v1.1           64.6 → 72.1     +11.6%<br>Agents' Last Exam         31.8 → 35.4     +11.3%<br>Automation-Bench v1.0.6   54.8 → 60.5     +10.4%<br>DeepSWE v1.1              74.2 → 77.0      +3.8%<br><br>Gains are concentrated in terminal operations and multi-step agent tasks; static knowledge benchmarks moved less. DeepSWE v1.1 at 77.0 edges out Claude Opus 5 at 74.0 and GPT-5.6 Sol at 73.0.<br><br>[WHERE IT FITS]<br>Agent and terminal automation — CLI operations, dependency installs, shell scripts, git workflows, log parsing. This is where the gains are largest.<br>Rapid prototyping from scratch — Describe what you want and get a runnable game, 3D scene, or interactive page. Good for validating an idea, not for polishing one.<br>Repository-level code changes — Multi-file refactors, cross-module rewrites, repo scaffolding. With 1M context you can load a mid-sized codebase in one shot.<br>Screenshot and design-to-code — Native multimodal, so images go in directly without wiring up a separate vision model.<br>High-volume batch work — The 35% discount plus shorter outputs suits workloads that run often and do not need top-tier output every time.<br><br>[PRICING]<br>HiLinkup charges 35% below DeepSeek's official rates, per 1M tokens.<br>Input (cache miss)     Official $0.15   →  HiLinkup $0.0975<br>Input (cache hit)      Official $0.003  →  HiLinkup $0.00195<br>Output                 Official $0.60   →  HiLinkup $0.39<br><br>DeepSeek uses peak/off-peak pricing, with peak at twice the off-peak rate ($0.30 / $0.006 / $1.20). HiLinkup applies the same 35% discount. The figures above are off-peak.<br><br>[QUICK START]<br>OpenAI-compatible. Change the model name; leave the rest of your code alone.<br><br>curl \"<a target=\"_blank\" rel=\"noopener noreferrer nofollow\" href=\"https://api.hilinkup.com/v1/chat/completions\">https://api.hilinkup.com/v1/chat/completions</a>\" \\</p><p>  -H \"Authorization: Bearer $HILINKUP_API_KEY\" \\</p><p>  -H \"Content-Type: application/json\" \\</p><p>  -d '{</p><p>  \"model\": \"doubao/deepseek-v4.1-flash-1017\",</p><p>  \"messages\": [</p><p>    {</p><p>      \"role\": \"user\",</p><p>      \"content\": \"Hello!\"</p><p>    }</p><p>  ]</p><p>}'</p>","category":"INFO","isPinned":false,"popup":false,"showBanner":false,"publishedAt":"2026-10-19T13:33:22.33061","id":"ann_mPJs8SKMsbXs-Zrw","title":"DeepSeek V4.1 Flash 1017 已上线，6.5折优惠","content":"<p>HiLinkup 已接入 DeepSeek V4.1 Flash 1017，现以官方价 6.5 折开放调用。<br><br>这不是一个新模型，是同一个月前发布的 V4.1 Flash 重新训练了一版权重。架构、参数量、上下文长度都没变，变的是后训练——跑分上去了，任务耗时和输出长度下来了，单个任务的实际花费随之下降。<br><br>【模型信息】<br>模型名称：DeepSeek V4.1 Flash 1017<br>模型 ID：doubao/deepseek-v4.1-flash-1017<br>上下文长度：1M tokens<br>最大输出：384K tokens<br>输入类型：文本 + 图像（原生多模态）<br>思考模式：支持，可调 reasoning effort<br><br>【这次升级改了什么】<br>1. 端到端项目生成能力增强<br>基于自反馈的强化学习训练，重点强化了从零构建完整项目的能力。在极少提示词、极少交互轮次下即可产出可运行的项目。官方演示案例包括故宫 3D 建模、网页版赛车游戏、Godot 引擎复刻《传送门 2》。<br><br>2. 输出更短，跑得更快<br>官方称之为性能与成本的帕累托改进。单价没有调整，但因为生成更少的 token，单任务账单明显下降。官方性能/成本图上，DeepSWE 单任务成本从约 0.50 美元降到约 0.20 美元。<br><br>3. 针对 Agent 平台定向优化<br>DeepSeek Harness、腾讯 CodeBuddy、WorkBuddy 均做了专项适配。<br><br>【能力提升（DeepSeek 官方数据）】<br>Terminal-Bench 4.0 31.2 → 42.3 +35.6%<br>ProgramBench 20.3 → 29.1 +43.3%<br>HLE（无工具） 36.8 → 42.6 +15.8%<br>Apex Bench v1.1 64.6 → 72.1 +11.6%<br>Agents' Last Exam 31.8 → 35.4 +11.3%<br>Automation-Bench v1.0.6 54.8 → 60.5 +10.4%<br>DeepSWE v1.1 74.2 → 77.0 +3.8%<br><br>涨幅集中在终端操作与多步 Agent 任务，静态知识型基准提升相对有限。DeepSWE v1.1 的 77.0 高于 Claude Opus 5 的 74.0 与 GPT-5.6 Sol 的 73.0。<br><br>【适合哪些场景】<br>Agent 与终端自动化：CLI 操作、依赖安装、shell 脚本、git 工作流、日志解析这类长链路任务，是本次提升最明显的方向。<br>从零搭原型：一句话描述需求，直接产出可跑的游戏、3D 场景、可视化页面。适合快速验证想法，而不是精雕细琢。<br>仓库级代码修改：多文件重构、跨模块改造、仓库结构生成。配合 1M 上下文可以直接吃下整个中型代码库。<br>截图与设计稿转代码：原生多模态，图像输入不需要额外拼接视觉模型。<br>高频批量调用：6.5 折的价格叠加更短的输出，适合跑量大、单次质量要求中等的任务。<br><br>【价格】<br>HiLinkup 定价为 DeepSeek 官方价的 6.5 折，按每 100 万 tokens 计。<br>输入（缓存未命中） 官方 0.15 美元 → HiLinkup 0.0975 美元<br>输入（缓存命中） 官方 0.003 美元 → HiLinkup 0.00195 美元<br>输出 官方 0.60 美元 → HiLinkup 0.39 美元<br><br>官方采用峰谷定价，峰时价为闲时价的 2 倍（0.30 / 0.006 / 1.20 美元）。HiLinkup 同样按 6.5 折折算，上表为闲时口径。<br><br>【如何调用】<br>OpenAI 兼容协议，只改 model 名，其余代码不用动。</p>"},{"announcementId":"ann_bwf8aZnWVo_hEGGw","titleZh":"领取2亿免费tokens","titleEn":"Claim 200 million free tokens","contentZh":"<p>活动期间充值至少 $10，自动激活2亿免费 Tokens 赠送权益。</p><p>免费 Tokens 仅限 DeepSeek-V4.1-Flash 与 GLM-5.3-Flash，其他模型不包含在内。</p><p>充值金额等额赠送，最高赠送 $20。</p>","contentMode":"RICH_TEXT","contentEn":"<p>Recharge during the campaign to activate 200M free Tokens. </p><p>The free Tokens are limited to DeepSeek-V4.1-Flash and GLM-5.3-Flash. Other models are not included. </p><p>The recharge bonus matches your recharge, capped at $20.</p>","category":"INFO","linkUrl":"https://hey.hilinkup.com/?ref=ref_N8eN6RDxB_9YumtW","isPinned":false,"popup":false,"showBanner":false,"publishedAt":"2026-10-16T15:12:10.020193","id":"ann_bwf8aZnWVo_hEGGw","title":"领取2亿免费tokens","content":"<p>活动期间充值至少 $10，自动激活2亿免费 Tokens 赠送权益。</p><p>免费 Tokens 仅限 DeepSeek-V4.1-Flash 与 GLM-5.3-Flash，其他模型不包含在内。</p><p>充值金额等额赠送，最高赠送 $20。</p>"}],"nextCursor":null,"hasMore":false,"limit":20},"error":null,"timestamp":1795233319909}