{"id":130430,"date":"2026-10-05T14:25:27","date_gmt":"2026-10-05T07:25:27","guid":{"rendered":"https:\/\/tino.vn\/blog\/?p=130430"},"modified":"2026-10-27T13:22:33","modified_gmt":"2026-10-27T06:22:33","slug":"toi-uu-chi-phi-api-cho-hermes-agent","status":"publish","type":"post","link":"https:\/\/tino.vn\/blog\/toi-uu-chi-phi-api-cho-hermes-agent\/","title":{"rendered":"H\u01b0\u1edbng d\u1eabn t\u1ed1i \u01b0u chi ph\u00ed API cho Hermes Agent b\u1eb1ng OpenRouter v\u00e0 Prompt Caching"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>\u0110\u1ec3 t\u1ed1i \u01b0u chi ph\u00ed API cho Hermes Agent, k\u1ebft h\u1ee3p OpenRouter v\u1edbi prompt caching: ph\u1ea7n prompt l\u1eb7p l\u1ea1i \u1edf m\u1ed7i l\u01b0\u1ee3t (system prompt, m\u00f4 t\u1ea3 c\u00f4ng c\u1ee5, b\u1ed9 nh\u1edb) \u0111\u01b0\u1ee3c \u0111\u1ecdc t\u1eeb cache v\u1edbi gi\u00e1 th\u1ea5p h\u01a1n nhi\u1ec1u so v\u1edbi input th\u01b0\u1eddng, v\u1edbi Claude l\u00e0 0,1 l\u1ea7n gi\u00e1 input.<\/strong> Ho\u00e1 \u0111\u01a1n gi\u1ea3m r\u00f5 khi agent ch\u1ea1y nhi\u1ec1u l\u01b0\u1ee3t li\u00ean t\u1ee5c.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">B\u00e0i thu\u1ed9c c\u1ee5m <a href=\"https:\/\/tino.vn\/blog\/nen-chon-llm-nao-cho-hermes-agent\/\">ch\u1ecdn LLM cho Hermes Agent<\/a> c\u1ee7a chuy\u00ean m\u1ee5c <a href=\"https:\/\/tino.vn\/blog\/hermes-agent-la-gi\/\">Hermes Agent<\/a>. B\u00e0i n\u00e0y gi\u1eef nguy\u00ean model b\u1ea1n \u0111ang d\u00f9ng v\u00e0 ch\u1ec9 c\u1ea5u h\u00ecnh cache; n\u1ebfu mu\u1ed1n \u0111\u1ed5i sang model r\u1ebb h\u01a1n, xem <a href=\"https:\/\/tino.vn\/blog\/llm-tiet-kiem-chi-phi-cho-hermes-agent\/\">LLM gi\u00e1 r\u1ebb cho Hermes Agent<\/a>. Ph\u1ea7n d\u01b0\u1edbi gi\u1ea3i th\u00edch kh\u00e1i ni\u1ec7m, c\u00e1c b\u01b0\u1edbc th\u1ef1c hi\u1ec7n v\u00e0 c\u00e1ch ki\u1ec3m tra k\u1ebft qu\u1ea3 tr\u00ean OpenRouter.<\/p>\n\n\n\n<h2 id=\"T\u1ea1i_sao_Hermes_Agent_t\u1ed1n_chi_ph\u00ed_API_h\u01a1n_chatbot_th\u00f4ng_th\u01b0\u1eddng?\">T\u1ea1i sao Hermes Agent t\u1ed1n chi ph\u00ed API h\u01a1n chatbot th\u00f4ng th\u01b0\u1eddng?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Hermes Agent t\u1ed1n h\u01a1n chatbot v\u00ec m\u1ed7i l\u01b0\u1ee3t g\u1ecdi model g\u1eedi k\u00e8m m\u1ed9t ph\u1ea7n prompt c\u1ed1 \u0111\u1ecbnh l\u1edbn, v\u00e0 m\u1ed9t t\u00e1c v\u1ee5 th\u01b0\u1eddng c\u1ea7n nhi\u1ec1u l\u01b0\u1ee3t. Tr\u01b0\u1edbc khi b\u1ea1n g\u00f5 ch\u1eef \u0111\u1ea7u ti\u00ean, agent \u0111\u00e3 \u0111\u00ednh k\u00e8m h\u01b0\u1edbng d\u1eabn h\u1ec7 th\u1ed1ng, danh s\u00e1ch c\u00f4ng c\u1ee5, danh m\u1ee5c k\u1ef9 n\u0103ng v\u00e0 c\u00e1c t\u1ec7p ghi nh\u1edb.<\/p>\n\n\n\n<h3 id=\"M\u1ed7i_l\u01b0\u1ee3t_ch\u1ea1y_\u0111\u1ec1u_g\u1eedi_l\u1ea1i_m\u1ed9t_&#8220;g\u00f3i_h\u00e0nh_l\u00fd&#8221;_l\u1edbn\">M\u1ed7i l\u01b0\u1ee3t ch\u1ea1y \u0111\u1ec1u g\u1eedi l\u1ea1i m\u1ed9t &#8220;g\u00f3i h\u00e0nh l\u00fd&#8221; l\u1edbn<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Khi chat v\u1edbi chatbot th\u01b0\u1eddng, h\u1ec7 th\u1ed1ng ch\u1ec9 g\u1eedi c\u00e2u h\u1ecfi v\u00e0 m\u1ed9t \u0111o\u1ea1n h\u01b0\u1edbng d\u1eabn ng\u1eafn. Hermes Agent th\u00ec g\u1eedi th\u00eam system prompt, m\u00f4 t\u1ea3 c\u00f4ng c\u1ee5 (tool schema) c\u1ee7a c\u00e1c toolset \u0111ang b\u1eadt, b\u1ed9 nh\u1edb v\u00e0 file ng\u1eef c\u1ea3nh. T\u00e0i li\u1ec7u Hermes g\u1ecdi ph\u1ea7n n\u00e0y l\u00e0 &#8220;fixed prompt budget&#8221;; \u0111o ph\u1ea7n c\u1ee7a b\u1ea1n b\u1eb1ng l\u1ec7nh d\u01b0\u1edbi (<a href=\"https:\/\/hermes-agent.nousresearch.com\/docs\/reference\/faq\" target=\"_blank\" rel=\"noopener nofollow\">Hermes Agent FAQ<\/a>):<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>hermes prompt-size<\/code><\/pre>\n\n\n\n<h3 id=\"V\u00f2ng_l\u1eb7p_c\u1ee7a_agent_nh\u00e2n_chi_ph\u00ed_l\u00ean_nhi\u1ec1u_l\u1ea7n\">V\u00f2ng l\u1eb7p c\u1ee7a agent nh\u00e2n chi ph\u00ed l\u00ean nhi\u1ec1u l\u1ea7n<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Agent kh\u00f4ng tr\u1ea3 l\u1eddi m\u1ed9t l\u1ea7n r\u1ed3i d\u1eebng. Hermes suy ngh\u0129, g\u1ecdi c\u00f4ng c\u1ee5, \u0111\u1ecdc k\u1ebft qu\u1ea3 r\u1ed3i suy ngh\u0129 ti\u1ebfp; m\u1ed7i v\u00f2ng g\u1eedi l\u1ea1i to\u00e0n b\u1ed9 &#8220;g\u00f3i h\u00e0nh l\u00fd&#8221;. M\u1ed9t t\u00e1c v\u1ee5 \u0111\u01a1n gi\u1ea3n c\u00f3 th\u1ec3 k\u00e9o theo nhi\u1ec1u l\u1ea7n g\u1ecdi API, n\u00ean chi ph\u00ed t\u0103ng theo s\u1ed1 v\u00f2ng m\u00e0 ng\u01b0\u1eddi d\u00f9ng kh\u00f3 nh\u1eadn ra.<\/p>\n\n\n\n<h2 id=\"Prompt_Caching_l\u00e0_g\u00ec?_C\u00f3_th\u1ec3_gi\u00fap_t\u1ed1i_\u01b0u_chi_ph\u00ed_API_ra_sao?\">Prompt Caching l\u00e0 g\u00ec? C\u00f3 th\u1ec3 gi\u00fap t\u1ed1i \u01b0u chi ph\u00ed API ra sao?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching l\u00e0 c\u01a1 ch\u1ebf l\u01b0u l\u1ea1i ph\u1ea7n \u0111\u1ea7u c\u1ee7a prompt \u0111\u00e3 x\u1eed l\u00fd, \u0111\u1ec3 c\u00e1c request sau c\u00f3 c\u00f9ng ph\u1ea7n \u0111\u1ea7u \u0111\u01b0\u1ee3c t\u00ednh gi\u00e1 \u0111\u1ecdc cache thay v\u00ec gi\u00e1 input \u0111\u1ea7y \u0111\u1ee7. \u0110i\u1ec1u ki\u1ec7n l\u00e0 ph\u1ea7n \u0111\u1ea7u \u0111\u00f3 ph\u1ea3i gi\u1ed1ng h\u1ec7t l\u1ea7n tr\u01b0\u1edbc v\u00e0 c\u00f2n trong th\u1eddi h\u1ea1n cache (TTL).<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Kh\u00f4ng c\u00f3 prompt caching:<\/strong> system prompt + tools + context + c\u00e2u h\u1ecfi m\u1edbi \u2192 t\u00ednh l\u1ea1i to\u00e0n b\u1ed9.<\/li>\n\n\n<li><strong>C\u00f3 prompt caching:<\/strong> ph\u1ea7n \u0111\u00e3 cache + c\u00e2u h\u1ecfi m\u1edbi \u2192 ph\u1ea7n \u0111\u00e3 cache t\u00ednh gi\u00e1 \u0111\u1ecdc cache, ch\u1ec9 ph\u1ea7n m\u1edbi t\u00ednh gi\u00e1 input.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">H\u00e3y h\u00ecnh dung b\u1ea1n n\u1ed9p m\u1ed9t t\u1eadp h\u1ed3 s\u01a1 d\u00e0y cho c\u00f9ng m\u1ed9t c\u01a1 quan nhi\u1ec1u l\u1ea7n trong ng\u00e0y. L\u1ea7n \u0111\u1ea7u n\u1ed9p \u0111\u1ee7 v\u00e0 tr\u1ea3 ph\u00ed l\u01b0u h\u1ed3 s\u01a1; t\u1eeb l\u1ea7n sau ch\u1ec9 c\u1ea7n n\u00f3i &#8220;d\u00f9ng l\u1ea1i t\u1eadp h\u1ed3 s\u01a1 h\u00f4m tr\u01b0\u1edbc&#8221; v\u00e0 tr\u1ea3 m\u1ed9t kho\u1ea3n nh\u1ecf.<\/p>\n\n\n\n<h2 id=\"Prompt_Caching_kh\u00e1c_Response_Caching_ra_sao?\">Prompt Caching kh\u00e1c Response Caching ra sao?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching l\u01b0u ph\u1ea7n \u0111\u1ea7u c\u1ee7a input, model v\u1eabn t\u1ea1o c\u00e2u tr\u1ea3 l\u1eddi m\u1edbi; response caching l\u01b0u nguy\u00ean c\u00e2u tr\u1ea3 l\u1eddi c\u1ee7a m\u1ed9t request gi\u1ed1ng h\u1ec7t v\u00e0 tr\u1ea3 l\u1ea1i m\u00e0 kh\u00f4ng g\u1ecdi model. Hai c\u01a1 ch\u1ebf \u0111\u1ed9c l\u1eadp v\u00e0 c\u00f3 th\u1ec3 d\u00f9ng c\u00f9ng l\u00fac.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Prompt Caching<\/th><th>Response Caching<\/th><\/tr><\/thead><tbody><tr><td>Cache m\u1ed9t ph\u1ea7n input gi\u1ed1ng nhau<\/td><td>Cache to\u00e0n b\u1ed9 request v\u00e0 response<\/td><\/tr><tr><td>Model v\u1eabn t\u1ea1o c\u00e2u tr\u1ea3 l\u1eddi m\u1edbi<\/td><td>Tr\u1ea3 l\u1ea1i response \u0111\u00e3 t\u1ea1o tr\u01b0\u1edbc \u0111\u00f3<\/td><\/tr><tr><td>H\u1ee3p h\u1ed9i tho\u1ea1i nhi\u1ec1u l\u01b0\u1ee3t<\/td><td>H\u1ee3p request gi\u1ed1ng h\u1ec7t nhau<\/td><\/tr><tr><td>Gi\u1ea3m gi\u00e1 input token<\/td><td>L\u1ea7n tr\u00fang cache kh\u00f4ng ph\u00e1t sinh token m\u1edbi<\/td><\/tr><tr><td>R\u1ea5t h\u1ee3p AI Agent<\/td><td>H\u1ee3p retry ho\u1eb7c t\u00e1c v\u1ee5 l\u1eb7p nguy\u00ean tr\u1ea1ng<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">OpenRouter h\u1ed7 tr\u1ee3 response caching qua header <code>X-OpenRouter-Cache<\/code>; cache g\u1eafn v\u1edbi t\u1eebng API key (<a href=\"https:\/\/openrouter.ai\/docs\/guides\/features\/response-caching\" target=\"_blank\" rel=\"noopener nofollow\">OpenRouter: response caching<\/a>). Hermes c\u00f3 s\u1eb5n c\u1ea5u h\u00ecnh cho c\u01a1 ch\u1ebf n\u00e0y v\u00e0 b\u1eadt m\u1eb7c \u0111\u1ecbnh (<a href=\"https:\/\/github.com\/NousResearch\/hermes-agent\/blob\/main\/cli-config.yaml.example\" target=\"_blank\" rel=\"noopener nofollow\">cli-config.yaml.example<\/a>).<\/p>\n\n\n\n<h2 id=\"C\u00e1ch_t\u1ed1i_\u01b0u_chi_ph\u00ed_API_cho_Hermes_Agent_b\u1eb1ng_OpenRouter_v\u00e0_Prompt_Caching\">C\u00e1ch t\u1ed1i \u01b0u chi ph\u00ed API cho Hermes Agent b\u1eb1ng OpenRouter v\u00e0 Prompt Caching<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Quy tr\u00ecnh g\u1ed3m t\u00e1m b\u01b0\u1edbc: k\u1ebft n\u1ed1i OpenRouter, ch\u1ecdn model h\u1ed7 tr\u1ee3 cache, \u0111\u1eb7t TTL, gi\u1eef model v\u00e0 ph\u1ea7n \u0111\u1ea7u prompt \u1ed5n \u0111\u1ecbnh, t\u1eadn d\u1ee5ng response caching, ki\u1ec3m tra k\u1ebft qu\u1ea3 tr\u00ean OpenRouter v\u00e0 \u0111\u1eb7t h\u1ea1n m\u1ee9c chi ti\u00eau. B\u01b0\u1edbc 3 (\u0111\u1eb7t TTL) v\u00e0 b\u01b0\u1edbc 7 (ki\u1ec3m cached_tokens) quy\u1ebft \u0111\u1ecbnh cache c\u00f3 gi\u1ea3m ho\u00e1 \u0111\u01a1n hay kh\u00f4ng.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_1:_K\u1ebft_n\u1ed1i_Hermes_Agent_v\u1edbi_OpenRouter\">B\u01b0\u1edbc 1: K\u1ebft n\u1ed1i Hermes Agent v\u1edbi OpenRouter<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Tino \u0111\u00e3 c\u00f3 h\u01b0\u1edbng d\u1eabn ri\u00eang cho ph\u1ea7n n\u00e0y: <a href=\"https:\/\/tino.vn\/blog\/ket-noi-hermes-agent-voi-openrouter-dung-gpt-va-claude\/\">k\u1ebft n\u1ed1i Hermes Agent v\u1edbi OpenRouter d\u00f9ng GPT v\u00e0 Claude<\/a>. Sau khi Hermes g\u1ecdi model th\u00e0nh c\u00f4ng qua OpenRouter, ti\u1ebfp t\u1ee5c t\u1ed1i \u01b0u prompt caching.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_2:_Ch\u1ecdn_model_ph\u00f9_h\u1ee3p_v\u1edbi_Prompt_Caching\">B\u01b0\u1edbc 2: Ch\u1ecdn model ph\u00f9 h\u1ee3p v\u1edbi Prompt Caching<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">N\u1ebfu mu\u1ed1n d\u00f9ng c\u01a1 ch\u1ebf prompt caching m\u00e0 Hermes \u0111i\u1ec1u khi\u1ec3n tr\u1ef1c ti\u1ebfp, Claude l\u00e0 l\u1ef1a ch\u1ecdn d\u1ec5 tri\u1ec3n khai: khi g\u1ecdi Claude qua OpenRouter ho\u1eb7c Anthropic API, Hermes \u0111\u1eb7t \u0111i\u1ec3m cache (<code>cache_control<\/code>) theo TTL b\u1ea1n c\u1ea5u h\u00ecnh. OpenAI, Grok, Moonshot, DeepSeek, Z.AI v\u00e0 Gemini 2.5 tr\u1edf l\u00ean t\u1ef1 cache \u1edf ph\u00eda nh\u00e0 cung c\u1ea5p, kh\u00f4ng c\u1ea7n <code>cache_control<\/code> (<a href=\"https:\/\/openrouter.ai\/docs\/guides\/best-practices\/prompt-caching\" target=\"_blank\" rel=\"noopener nofollow\">OpenRouter: prompt caching<\/a>).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>L\u01b0u \u00fd:<\/strong> kh\u00f4ng n\u00ean ch\u1ecdn model ch\u1ec9 d\u1ef1a tr\u00ean m\u1ee9c gi\u1ea3m khi cache. C\u1ea7n c\u00e2n th\u00eam:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Gi\u00e1 input.<\/li>\n\n\n<li>Gi\u00e1 output.<\/li>\n\n\n<li>Ch\u1ea5t l\u01b0\u1ee3ng ph\u1ea3n h\u1ed3i.<\/li>\n\n\n<li>Kh\u1ea3 n\u0103ng g\u1ecdi tool.<\/li>\n\n\n<li>Context window.<\/li>\n\n\n<li>T\u1ed1c \u0111\u1ed9 ph\u1ea3n h\u1ed3i.<\/li>\n\n\n<li>M\u1ee9c \u0111\u1ed9 ph\u00f9 h\u1ee3p v\u1edbi kh\u1ed1i l\u01b0\u1ee3ng vi\u1ec7c.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Model r\u1ebb h\u01a1n nh\u01b0ng t\u1ea1o output qu\u00e1 d\u00e0i ho\u1eb7c c\u1ea7n nhi\u1ec1u v\u00f2ng s\u1eeda ch\u01b0a ch\u1eafc cho t\u1ed5ng chi ph\u00ed th\u1ea5p h\u01a1n. Prompt qu\u00e1 ng\u1eafn c\u0169ng kh\u00f4ng \u0111\u01b0\u1ee3c cache: OpenRouter ghi m\u1ee9c t\u1ed1i thi\u1ec3u kho\u1ea3ng 1.024 token v\u1edbi OpenAI, t\u1eeb 512 \u0111\u1ebfn 4.096 token v\u1edbi Anthropic tu\u1ef3 model.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_3:_C\u1ea5u_h\u00ecnh_TTL_cho_Prompt_Caching\">B\u01b0\u1edbc 3: C\u1ea5u h\u00ecnh TTL cho Prompt Caching<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Hermes Agent h\u1ed7 tr\u1ee3 ba gi\u00e1 tr\u1ecb:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>prompt_caching:\n  cache_ttl: \"5m\"<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">ho\u1eb7c:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>prompt_caching:\n  cache_ttl: \"1h\"<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">ho\u1eb7c:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>prompt_caching:\n  cache_ttl: \"auto\"<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">C\u1ea5u h\u00ecnh n\u1eb1m trong file <code>~\/.hermes\/config.yaml<\/code>; m\u1edf file b\u1eb1ng:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>hermes config edit<\/code><\/pre>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1024\" height=\"596\" src=\"https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/1-2-1024x596.png\" alt=\"File config.yaml c\u1ee7a Hermes Agent m\u1edf b\u1eb1ng nano, kh\u1ed1i prompt_caching v\u1edbi cache_ttl 5m\" class=\"wp-image-130434\" title=\"\" srcset=\"https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/1-2-1024x596.png 1024w, https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/1-2-300x174.png 300w, https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/1-2-767x446.png 767w, https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/1-2.png 1145w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Theo t\u00e0i li\u1ec7u Configuration, ch\u1ebf \u0111\u1ed9 <strong>auto<\/strong> ch\u1ecdn TTL theo lo\u1ea1i phi\u00ean (<a href=\"https:\/\/hermes-agent.nousresearch.com\/docs\/user-guide\/configuration\" target=\"_blank\" rel=\"noopener nofollow\">Hermes Agent: Configuration<\/a>, 26\/10\/2026):<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Phi\u00ean c\u00f3 ng\u01b0\u1eddi g\u00f5 tr\u1ef1c ti\u1ebfp d\u00f9ng TTL 1 gi\u1edd.<\/li>\n\n\n<li>Phi\u00ean ch\u1ea1y t\u1ef1 \u0111\u1ed9ng (cron, webhook, API, batch) d\u00f9ng TTL 5 ph\u00fat; subagent lu\u00f4n d\u00f9ng 5 ph\u00fat.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Gi\u00e1 tr\u1ecb kh\u00e1c &#8220;5m&#8221;, &#8220;1h&#8221;, &#8220;auto&#8221; b\u1ecb b\u1ecf qua. V\u1edbi ph\u1ea7n l\u1edbn ng\u01b0\u1eddi d\u00f9ng, <strong>auto<\/strong> l\u00e0 l\u1ef1a ch\u1ecdn h\u1ee3p l\u00fd v\u00ec kh\u00f4ng c\u1ea7n t\u1ef1 \u0111\u1ed5i TTL gi\u1eefa c\u00e1c lo\u1ea1i vi\u1ec7c. Sau khi ch\u1ec9nh c\u1ea5u h\u00ecnh, m\u1edf phi\u00ean Hermes m\u1edbi ho\u1eb7c kh\u1edfi \u0111\u1ed9ng l\u1ea1i gateway \u0111\u1ec3 c\u1ea5u h\u00ecnh m\u1edbi \u0111\u01b0\u1ee3c n\u1ea1p.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_4:_Gi\u1eef_model_v\u00e0_phi\u00ean_l\u00e0m_vi\u1ec7c_\u1ed5n_\u0111\u1ecbnh\">B\u01b0\u1edbc 4: Gi\u1eef model v\u00e0 phi\u00ean l\u00e0m vi\u1ec7c \u1ed5n \u0111\u1ecbnh<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching hi\u1ec7u qu\u1ea3 khi c\u00e1c request sau v\u1eabn d\u00f9ng c\u00f9ng model v\u00e0 c\u00f9ng nh\u00e0 cung c\u1ea5p. OpenRouter t\u1ef1 gi\u1eef request ti\u1ebfp theo \u1edf c\u00f9ng nh\u00e0 cung c\u1ea5p (sticky routing) \u0111\u1ec3 cache c\u00f2n d\u00f9ng \u0111\u01b0\u1ee3c; phi\u00ean im l\u1eb7ng qu\u00e1 10 ph\u00fat th\u00ec h\u1ebft gi\u1eef tuy\u1ebfn, v\u00e0 sticky routing b\u1ecb b\u1ecf qua n\u1ebfu b\u1ea1n \u0111\u1eb7t th\u1ee9 t\u1ef1 nh\u00e0 cung c\u1ea5p th\u1ee7 c\u00f4ng (<a href=\"https:\/\/openrouter.ai\/docs\/guides\/best-practices\/prompt-caching\" target=\"_blank\" rel=\"noopener nofollow\">OpenRouter: prompt caching<\/a>).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">M\u1ed9t s\u1ed1 thao t\u00e1c l\u00e0m m\u1ea5t cache hi\u1ec7n t\u1ea1i:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u0110\u1ed5i model gi\u1eefa phi\u00ean.<\/li>\n\n\n<li>Chuy\u1ec3n provider.<\/li>\n\n\n<li>Fallback sang provider kh\u00e1c.<\/li>\n\n\n<li>Chuy\u1ec3n sang API key kh\u00e1c.<\/li>\n\n\n<li>Cache h\u1ebft TTL.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">T\u00e0i li\u1ec7u Hermes l\u01b0u \u00fd \u0111\u1ed5i model b\u1eb1ng <code>\/model<\/code> l\u00e0m m\u1ea5t prompt cache, n\u00ean l\u01b0\u1ee3t k\u1ebf ti\u1ebfp \u0111\u1ecdc l\u1ea1i context \u1edf gi\u00e1 input \u0111\u1ea7y \u0111\u1ee7. V\u00ec v\u1eady kh\u00f4ng n\u00ean \u0111\u1ed5i model li\u00ean t\u1ee5c \u0111\u1ec3 ti\u1ebft ki\u1ec7m v\u00e0i request ng\u1eafn khi phi\u00ean \u0111\u00e3 c\u00f3 context l\u1edbn.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_5:_Gi\u1eef_ph\u1ea7n_prompt_d\u00f9ng_chung_c\u00e0ng_\u1ed5n_\u0111\u1ecbnh_c\u00e0ng_t\u1ed1t\">B\u01b0\u1edbc 5: Gi\u1eef ph\u1ea7n prompt d\u00f9ng chung c\u00e0ng \u1ed5n \u0111\u1ecbnh c\u00e0ng t\u1ed1t<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching ch\u1ec9 ho\u1ea1t \u0111\u1ed9ng khi ph\u1ea7n \u0111\u1ea7u prompt gi\u1ed1ng h\u1ec7t gi\u1eefa c\u00e1c l\u01b0\u1ee3t. Ph\u1ea7n \u0111\u1ea7u n\u00ean ch\u1ee9a n\u1ed9i dung \u00edt thay \u0111\u1ed5i:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>System prompt.<\/li>\n\n\n<li>Tool definitions.<\/li>\n\n\n<li>Schema.<\/li>\n\n\n<li>H\u01b0\u1edbng d\u1eabn c\u1ed1 \u0111\u1ecbnh.<\/li>\n\n\n<li>Skill.<\/li>\n\n\n<li>T\u00e0i li\u1ec7u tham chi\u1ebfu d\u00f9ng xuy\u00ean su\u1ed1t phi\u00ean.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>N\u1ed9i dung thay \u0111\u1ed5i th\u01b0\u1eddng xuy\u00ean n\u00ean n\u1eb1m v\u1ec1 sau:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>C\u00e2u h\u1ecfi m\u1edbi.<\/li>\n\n\n<li>Tool result m\u1edbi.<\/li>\n\n\n<li>Timestamp.<\/li>\n\n\n<li>Tr\u1ea1ng th\u00e1i t\u1ea1m th\u1eddi.<\/li>\n\n\n<li>D\u1eef li\u1ec7u ph\u00e1t sinh trong t\u1eebng l\u01b0\u1ee3t.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">N\u1ebfu ph\u1ea7n \u0111\u1ea7u request li\u00ean t\u1ee5c thay \u0111\u1ed5i, kh\u1ea3 n\u0103ng tr\u00fang cache gi\u1ea3m. <strong>Ngo\u00e0i ra, b\u1ea1n n\u00ean:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Kh\u00f4ng b\u1eadt\/t\u1eaft c\u00f4ng c\u1ee5 gi\u1eefa phi\u00ean.<\/li>\n\n\n<li>Kh\u00f4ng s\u1eeda t\u1ec7p ghi nh\u1edb khi \u0111ang ch\u1ea1y t\u00e1c v\u1ee5 d\u00e0i.<\/li>\n\n\n<li>Gi\u1eef kho\u1ea3ng c\u00e1ch gi\u1eefa c\u00e1c l\u01b0\u1ee3t d\u01b0\u1edbi TTL (5 ph\u00fat ho\u1eb7c 1 gi\u1edd).<\/li>\n<\/ul>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_6:_T\u1eadn_d\u1ee5ng_Response_Caching_c\u1ee7a_OpenRouter\">B\u01b0\u1edbc 6: T\u1eadn d\u1ee5ng Response Caching c\u1ee7a OpenRouter<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Ngo\u00e0i prompt caching, Hermes Agent h\u1ed7 tr\u1ee3 <strong>OpenRouter Response Caching<\/strong>. C\u1ea5u h\u00ecnh t\u01b0\u01a1ng \u1ee9ng:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>openrouter:\n  response_cache: true\n  response_cache_ttl: 300<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Trong \u0111\u00f3 300 gi\u00e2y = 5 ph\u00fat. Theo file c\u1ea5u h\u00ecnh m\u1eabu c\u1ee7a Hermes, <code>response_cache<\/code> m\u1eb7c \u0111\u1ecbnh l\u00e0 <code>true<\/code>, TTL m\u1eb7c \u0111\u1ecbnh 300 gi\u00e2y, \u0111\u1eb7t \u0111\u01b0\u1ee3c t\u1eeb 1 \u0111\u1ebfn 86.400 gi\u00e2y.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Response caching h\u1eefu \u00edch khi:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Agent retry c\u00f9ng request.<\/li>\n\n\n<li>Workflow b\u1ecb ch\u1ea1y l\u1ea1i.<\/li>\n\n\n<li>M\u1ed9t t\u00e1c v\u1ee5 h\u1ecfi c\u00f9ng d\u1eef li\u1ec7u nhi\u1ec1u l\u1ea7n.<\/li>\n\n\n<li>API client g\u1eedi request tr\u00f9ng nhau.<\/li>\n\n\n<li>M\u1ed9t b\u01b0\u1edbc trong pipeline b\u1ecb th\u1ef1c thi l\u1ea1i.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">N\u1ebfu request gi\u1ed1ng ho\u00e0n to\u00e0n v\u00e0 cache c\u00f2n hi\u1ec7u l\u1ef1c, OpenRouter tr\u1ea3 l\u1ea1i response \u0111\u00e3 l\u01b0u m\u00e0 kh\u00f4ng ph\u00e1t sinh l\u01b0\u1ee3t suy lu\u1eadn m\u1edbi. Response caching v\u00e0 prompt caching v\u1eabn l\u00e0 hai c\u01a1 ch\u1ebf \u0111\u1ed9c l\u1eadp.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_7:_Ki\u1ec3m_tra_Prompt_Caching_c\u00f3_th\u1eadt_s\u1ef1_ti\u1ebft_ki\u1ec7m_chi_ph\u00ed_kh\u00f4ng\">B\u01b0\u1edbc 7: Ki\u1ec3m tra Prompt Caching c\u00f3 th\u1eadt s\u1ef1 ti\u1ebft ki\u1ec7m chi ph\u00ed kh\u00f4ng<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Kh\u00f4ng n\u00ean b\u1eadt cache r\u1ed3i gi\u1ea3 \u0111\u1ecbnh chi ph\u00ed \u0111\u00e3 gi\u1ea3m. Ki\u1ec3m tra tr\u1ef1c ti\u1ebfp tr\u00ean trang Activity trong t\u00e0i kho\u1ea3n OpenRouter. Trong th\u00f4ng tin usage c\u1ee7a request, tr\u01b0\u1eddng <strong>cached_tokens<\/strong> cho bi\u1ebft l\u01b0\u1ee3ng token \u0111\u1ecdc t\u1eeb cache.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"704\" height=\"426\" src=\"https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/2-2.png\" alt=\"Bi\u1ec3u \u0111\u1ed3 Prompt token caching tr\u00ean OpenRouter: ph\u1ea7n token cached v\u00e0 uncached theo ng\u00e0y\" class=\"wp-image-130433\" title=\"\" srcset=\"https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/2-2.png 704w, https:\/\/tino.vn\/blog\/wp-content\/uploads\/2026\/10\/2-2-300x182.png 300w\" sizes=\"(max-width: 704px) 100vw, 704px\" \/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">OpenRouter c\u00f2n c\u00f3 <strong>cache_write_tokens<\/strong> cho bi\u1ebft l\u01b0\u1ee3ng token ghi v\u00e0o cache. N\u1ebfu <strong>cached_tokens &gt; 0<\/strong>, request \u0111\u00e3 tr\u00fang cache. Tr\u01b0\u1eddng <strong>cache_discount<\/strong> cho bi\u1ebft cache \u0111\u00e3 ti\u1ebft ki\u1ec7m bao nhi\u00eau cho request \u0111\u00f3. V\u1edbi Claude qua OpenRouter, Hermes CLI hi\u1ec7n d\u00f2ng &#8220;Prompt caching: enabled&#8221; khi c\u01a1 ch\u1ebf n\u00e0y \u0111\u01b0\u1ee3c b\u1eadt.<\/p>\n\n\n\n<h3 id=\"B\u01b0\u1edbc_8:_\u0110\u1eb7t_h\u1ea1n_m\u1ee9c_chi_ti\u00eau_\u0111\u1ec3_an_t\u00e2m\">B\u01b0\u1edbc 8: \u0110\u1eb7t h\u1ea1n m\u1ee9c chi ti\u00eau \u0111\u1ec3 an t\u00e2m<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">D\u00f9 \u0111\u00e3 t\u1ed1i \u01b0u, agent v\u1eabn c\u00f3 th\u1ec3 r\u01a1i v\u00e0o v\u00f2ng l\u1eb7p b\u1ea5t ng\u1edd. N\u1ea1p s\u1ed1 ti\u1ec1n v\u1eeba ph\u1ea3i v\u00e0 \u0111\u1eb7t gi\u1edbi h\u1ea1n credit cho t\u1eebng API key trong ph\u1ea7n qu\u1ea3n l\u00fd key c\u1ee7a OpenRouter \u0111\u1ec3 chi ph\u00ed kh\u00f4ng v\u01b0\u1ee3t t\u1ea7m ki\u1ec3m so\u00e1t.<\/p>\n\n\n\n<h2 id=\"Prompt_Caching_c\u00f3_th\u1ec3_ti\u1ebft_ki\u1ec7m_bao_nhi\u00eau_chi_ph\u00ed?\">Prompt Caching c\u00f3 th\u1ec3 ti\u1ebft ki\u1ec7m bao nhi\u00eau chi ph\u00ed?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">V\u1edbi Claude, \u0111\u1ecdc cache t\u00ednh 0,1 l\u1ea7n gi\u00e1 input, ghi cache 5 ph\u00fat t\u00ednh 1,25 l\u1ea7n v\u00e0 ghi cache 1 gi\u1edd t\u00ednh 2 l\u1ea7n (<a href=\"https:\/\/openrouter.ai\/docs\/guides\/best-practices\/prompt-caching\" target=\"_blank\" rel=\"noopener nofollow\">OpenRouter: prompt caching<\/a>). V\u00ed d\u1ee5 v\u1edbi Claude Sonnet 4.6, model trong \u1ea3nh c\u1ea5u h\u00ecnh \u1edf b\u01b0\u1edbc 3:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Lo\u1ea1i token<\/th><th>Chi ph\u00ed tham kh\u1ea3o<\/th><\/tr><\/thead><tbody><tr><td>Input th\u00f4ng th\u01b0\u1eddng<\/td><td>3 USD \/ 1 tri\u1ec7u token<\/td><\/tr><tr><td>\u0110\u1ecdc cache<\/td><td>0,30 USD \/ 1 tri\u1ec7u token<\/td><\/tr><tr><td>Ghi cache 5 ph\u00fat<\/td><td>3,75 USD \/ 1 tri\u1ec7u token<\/td><\/tr><tr><td>Ghi cache 1 gi\u1edd<\/td><td>6 USD \/ 1 tri\u1ec7u token<\/td><\/tr><\/tbody><\/table><figcaption class=\"wp-element-caption\">Theo danh m\u1ee5c model OpenRouter ng\u00e0y 26\/10\/2026. Claude Sonnet 5.5 hi\u1ec7n c\u00f3 gi\u00e1 2 \/ 0,10 \/ 2,5 \/ 4 USD cho c\u00f9ng b\u1ed1n lo\u1ea1i token.<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Gi\u1ea3 s\u1eed Hermes Agent c\u00f3 kho\u1ea3ng <strong>10.000 token context l\u1eb7p l\u1ea1i<\/strong> trong 6 l\u01b0\u1ee3t li\u00ean ti\u1ebfp v\u1edbi Claude Sonnet 4.6.<\/p>\n\n\n\n<h3 id=\"Kh\u00f4ng_d\u00f9ng_cache\">Kh\u00f4ng d\u00f9ng cache<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>6 l\u01b0\u1ee3t \u00d7 10.000 token = 60.000 token\n60.000 \/ 1.000.000 \u00d7 3 USD = 0,18 USD<\/code><\/pre>\n\n\n\n<h3 id=\"D\u00f9ng_Prompt_Caching_5_ph\u00fat\">D\u00f9ng Prompt Caching 5 ph\u00fat<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">L\u01b0\u1ee3t \u0111\u1ea7u ghi 10.000 token v\u00e0o cache, 5 l\u01b0\u1ee3t sau \u0111\u1ecdc t\u1eeb cache:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>10.000 \/ 1.000.000 \u00d7 3,75 USD = 0,0375 USD\n50.000 \/ 1.000.000 \u00d7 0,30 USD = 0,015 USD\nT\u1ed5ng: 0,0375 + 0,015 = 0,0525 USD<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">So v\u1edbi 0,18 USD khi kh\u00f4ng cache, ph\u1ea7n context l\u1eb7p l\u1ea1i trong v\u00ed d\u1ee5 n\u00e0y gi\u1ea3m kho\u1ea3ng <strong>71% chi ph\u00ed input<\/strong>. \u0110\u00e2y ch\u1ec9 l\u00e0 v\u00ed d\u1ee5 minh ho\u1ea1; ho\u00e1 \u0111\u01a1n th\u1ef1c t\u1ebf c\u00f2n ph\u1ee5 thu\u1ed9c:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Token m\u1edbi \u1edf m\u1ed7i l\u01b0\u1ee3t.<\/li>\n\n\n<li>Output token.<\/li>\n\n\n<li>Tool calls.<\/li>\n\n\n<li>Model v\u00e0 provider.<\/li>\n\n\n<li>TTL.<\/li>\n\n\n<li>S\u1ed1 l\u1ea7n cache \u0111\u01b0\u1ee3c t\u00e1i s\u1eed d\u1ee5ng.<\/li>\n\n\n<li>Gi\u00e1 API t\u1ea1i th\u1eddi \u0111i\u1ec3m s\u1eed d\u1ee5ng.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching c\u00f3 l\u1ee3i r\u00f5 khi m\u1ed9t l\u01b0\u1ee3ng context l\u1edbn \u0111\u01b0\u1ee3c d\u00f9ng l\u1ea1i nhi\u1ec1u l\u1ea7n. V\u00ed d\u1ee5 t\u00ednh chi ph\u00ed c\u1ea3 th\u00e1ng cho 8 model, c\u00f3 v\u00e0 kh\u00f4ng cache, xem b\u00e0i <a href=\"https:\/\/tino.vn\/blog\/hermes-agent-1-thang-ton-bao-nhieu-chi-phi\/\">Hermes Agent 1 th\u00e1ng t\u1ed1n bao nhi\u00eau chi ph\u00ed<\/a>.<\/p>\n\n\n\n<h2 id=\"N\u00ean_ch\u1ecdn_TTL_5_ph\u00fat,_1_gi\u1edd_hay_Auto?\">N\u00ean ch\u1ecdn TTL 5 ph\u00fat, 1 gi\u1edd hay Auto?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Ch\u1ecdn TTL theo kho\u1ea3ng c\u00e1ch gi\u1eefa c\u00e1c l\u01b0\u1ee3t: l\u01b0\u1ee3t li\u00ean t\u1ee5c d\u01b0\u1edbi 5 ph\u00fat th\u00ec 5 ph\u00fat l\u00e0 \u0111\u1ee7 v\u00e0 ghi cache r\u1ebb h\u01a1n; ng\u01b0\u1eddi d\u00f9ng hay ngh\u1ec9 v\u00e0i ph\u00fat gi\u1eefa c\u00e1c l\u01b0\u1ee3t th\u00ec 1 gi\u1edd tr\u00fang cache nhi\u1ec1u h\u01a1n; kh\u00f4ng mu\u1ed1n t\u1ef1 qu\u1ea3n l\u00fd th\u00ec \u0111\u1ec3 auto.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Ki\u1ec3u s\u1eed d\u1ee5ng<\/th><th>TTL ph\u00f9 h\u1ee3p<\/th><\/tr><\/thead><tbody><tr><td>Chat li\u00ean t\u1ee5c<\/td><td><code>5m<\/code><\/td><\/tr><tr><td>Agent loop nhanh<\/td><td><code>5m<\/code><\/td><\/tr><tr><td>Cron job<\/td><td><code>5m<\/code><\/td><\/tr><tr><td>Webhook<\/td><td><code>5m<\/code><\/td><\/tr><tr><td>Batch\/API t\u1ef1 \u0111\u1ed9ng<\/td><td><code>5m<\/code><\/td><\/tr><tr><td>Ng\u01b0\u1eddi d\u00f9ng chat v\u00e0 ngh\u1ec9 v\u00e0i ph\u00fat<\/td><td><code>1h<\/code><\/td><\/tr><tr><td>Phi\u00ean nghi\u00ean c\u1ee9u d\u00e0i<\/td><td><code>1h<\/code><\/td><\/tr><tr><td>Kh\u00f4ng mu\u1ed1n t\u1ef1 qu\u1ea3n l\u00fd<\/td><td><code>auto<\/code><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>L\u01b0u \u00fd:<\/strong> TTL d\u00e0i h\u01a1n kh\u00f4ng ph\u1ea3i l\u00fac n\u00e0o c\u0169ng r\u1ebb h\u01a1n, v\u00ec ghi cache 1 gi\u1edd \u0111\u1eaft h\u01a1n ghi cache 5 ph\u00fat.<\/p>\n\n\n\n<h2 id=\"Nh\u1eefng_nguy\u00ean_nh\u00e2n_khi\u1ebfn_Prompt_Caching_kh\u00f4ng_ti\u1ebft_ki\u1ec7m_nh\u01b0_mong_\u0111\u1ee3i\">Nh\u1eefng nguy\u00ean nh\u00e2n khi\u1ebfn Prompt Caching kh\u00f4ng ti\u1ebft ki\u1ec7m nh\u01b0 mong \u0111\u1ee3i<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Cache kh\u00f4ng ti\u1ebft ki\u1ec7m khi ph\u1ea7n \u0111\u1ea7u prompt kh\u00f4ng \u0111\u01b0\u1ee3c d\u00f9ng l\u1ea1i: qu\u00e1 ng\u1eafn \u0111\u1ec3 cache, h\u1ebft TTL tr\u01b0\u1edbc l\u01b0\u1ee3t sau, ho\u1eb7c b\u1ecb \u0111\u1ed5i model, API key, nh\u00e0 cung c\u1ea5p. Khi \u0111\u00f3 b\u1ea1n v\u1eabn tr\u1ea3 ph\u00ed ghi cache m\u00e0 kh\u00f4ng \u0111\u01b0\u1ee3c gi\u1ea3m \u1edf l\u01b0\u1ee3t sau. C\u00e1c tr\u01b0\u1eddng h\u1ee3p th\u01b0\u1eddng g\u1eb7p:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Prompt qu\u00e1 ng\u1eafn:<\/strong> ph\u1ea7n context ch\u01b0a \u0111\u1ee7 l\u1edbn \u0111\u1ec3 nh\u00e0 cung c\u1ea5p t\u1ea1o cache.<\/li>\n\n\n<li><strong>TTL h\u1ebft h\u1ea1n:<\/strong> request ti\u1ebfp theo \u0111\u1ebfn qu\u00e1 mu\u1ed9n.<\/li>\n\n\n<li><strong>\u0110\u1ed5i model gi\u1eefa phi\u00ean:<\/strong> cache c\u1ee7a model c\u0169 kh\u00f4ng d\u00f9ng l\u1ea1i \u0111\u01b0\u1ee3c.<\/li>\n\n\n<li><strong>\u0110\u1ed5i API key:<\/strong> cache c\u00f3 th\u1ec3 g\u1eafn v\u1edbi t\u00e0i kho\u1ea3n ho\u1eb7c credential \u0111ang d\u00f9ng.<\/li>\n\n\n<li><strong>Provider thay \u0111\u1ed5i:<\/strong> request \u0111\u01b0\u1ee3c chuy\u1ec3n sang endpoint ch\u01b0a c\u00f3 cache.<\/li>\n\n\n<li><strong>Prefix thay \u0111\u1ed5i:<\/strong> system prompt ho\u1eb7c ph\u1ea7n \u0111\u1ea7u context b\u1ecb ch\u1ec9nh th\u01b0\u1eddng xuy\u00ean.<\/li>\n\n\n<li><strong>Ch\u1ec9 g\u1eedi m\u1ed9t request:<\/strong> ph\u00ed ghi cache c\u00f3 th\u1ec3 cao h\u01a1n l\u1ee3i \u00edch.<\/li>\n\n\n<li><strong>Context li\u00ean t\u1ee5c thay \u0111\u1ed5i:<\/strong> t\u1ef7 l\u1ec7 token \u0111\u1ecdc \u0111\u01b0\u1ee3c t\u1eeb cache th\u1ea5p.<\/li>\n<\/ul>\n\n\n\n<h2 id=\"M\u1ed9t_s\u1ed1_c\u00e1ch_gi\u1ea3m_th\u00eam_chi_ph\u00ed_API_cho_Hermes_Agent\">M\u1ed9t s\u1ed1 c\u00e1ch gi\u1ea3m th\u00eam chi ph\u00ed API cho Hermes Agent<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching ch\u1ec9 l\u00e0 m\u1ed9t ph\u1ea7n c\u1ee7a b\u00e0i to\u00e1n. K\u1ebft h\u1ee3p th\u00eam vi\u1ec7c chia model theo \u0111\u1ed9 kh\u00f3, b\u1edbt toolset, n\u00e9n h\u1ed9i tho\u1ea1i d\u00e0i v\u00e0 \u0111\u1eb7t h\u1ea1n m\u1ee9c \u0111\u1ec3 gi\u1ea3m ti\u1ebfp ph\u1ea7n token ph\u1ea3i tr\u1ea3. C\u00e1c c\u00e1ch d\u01b0\u1edbi kh\u00f4ng \u0111\u00f2i \u0111\u1ed5i model ch\u00ednh, n\u00ean \u00e1p d\u1ee5ng \u0111\u01b0\u1ee3c ngay c\u1ea3 khi b\u1ea1n \u0111ang h\u00e0i l\u00f2ng v\u1edbi ch\u1ea5t l\u01b0\u1ee3ng hi\u1ec7n t\u1ea1i:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>D\u00f9ng model m\u1ea1nh cho nhi\u1ec7m v\u1ee5 ph\u1ee9c t\u1ea1p, model r\u1ebb h\u01a1n cho t\u00e1c v\u1ee5 \u0111\u01a1n gi\u1ea3n.<\/li>\n\n\n<li>H\u1ea1n ch\u1ebf \u0111\u1ed5i model gi\u1eefa m\u1ed9t phi\u00ean context l\u1edbn.<\/li>\n\n\n<li>Kh\u00f4ng \u0111\u01b0a t\u00e0i li\u1ec7u d\u00e0i v\u00e0o prompt khi kh\u00f4ng th\u1eadt s\u1ef1 c\u1ea7n.<\/li>\n\n\n<li>Chia Skill theo t\u1eebng m\u1ee5c \u0111\u00edch thay v\u00ec t\u1ea3i qu\u00e1 nhi\u1ec1u h\u01b0\u1edbng d\u1eabn c\u00f9ng l\u00fac.<\/li>\n\n\n<li>D\u00f9ng <code>\/compress<\/code> ho\u1eb7c \u0111\u1ec3 Hermes t\u1ef1 n\u00e9n khi h\u1ed9i tho\u1ea1i d\u00e0i.<\/li>\n\n\n<li>D\u00f9ng model gi\u00e1 th\u1ea5p cho vi\u1ec7c ph\u1ee5 nh\u01b0 n\u00e9n context qua kho\u00e1 <code>auxiliary<\/code>.<\/li>\n\n\n<li>Ki\u1ec3m tra OpenRouter Activity \u0111\u1ec3 t\u00ecm workflow ti\u00eau th\u1ee5 token b\u1ea5t th\u01b0\u1eddng.<\/li>\n\n\n<li>\u0110\u1eb7t gi\u1edbi h\u1ea1n ng\u00e2n s\u00e1ch API.<\/li>\n\n\n<li>Gi\u1edbi h\u1ea1n \u0111\u1ed9 d\u00e0i output n\u1ebfu kh\u00f4ng c\u1ea7n ph\u1ea3n h\u1ed3i d\u00e0i.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Hermes c\u00f2n h\u1ed7 tr\u1ee3 <code>provider_routing<\/code> \u0111\u1ec3 ch\u1ecdn c\u00e1ch OpenRouter x\u1ebfp nh\u00e0 cung c\u1ea5p. Theo file c\u1ea5u h\u00ecnh m\u1eabu, <code>sort: \"price\"<\/code> l\u00e0 m\u1eb7c \u0111\u1ecbnh; \u0111\u1ed5i sang <code>\"throughput\"<\/code> ho\u1eb7c <code>\"latency\"<\/code> khi \u01b0u ti\u00ean t\u1ed1c \u0111\u1ed9 (<a href=\"https:\/\/github.com\/NousResearch\/hermes-agent\/blob\/main\/cli-config.yaml.example\" target=\"_blank\" rel=\"noopener nofollow\">cli-config.yaml.example<\/a>).<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>provider_routing:\n  sort: \"price\"<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">V\u1edbi h\u1ed9i tho\u1ea1i d\u00e0i \u0111\u00e3 c\u00f3 cache t\u1ed1t, gi\u1eef tuy\u1ebfn \u1ed5n \u0111\u1ecbnh \u0111\u00f4i khi quan tr\u1ecdng h\u01a1n c\u1ed1 gi\u1ea3m m\u1ed9t kho\u1ea3n r\u1ea5t nh\u1ecf tr\u00ean t\u1eebng request.<\/p>\n\n\n\n<h2 id=\"Khi_n\u00e0o_kh\u00f4ng_c\u1ea7n_qu\u00e1_quan_t\u00e2m_\u0111\u1ebfn_Prompt_Caching?\">Khi n\u00e0o kh\u00f4ng c\u1ea7n qu\u00e1 quan t\u00e2m \u0111\u1ebfn Prompt Caching?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt caching \u00edt t\u1ea1o kh\u00e1c bi\u1ec7t khi m\u1ed7i t\u00e1c v\u1ee5 ch\u1ec9 c\u00f3 m\u1ed9t request ng\u1eafn, context \u0111\u1ed5i ho\u00e0n to\u00e0n gi\u1eefa c\u00e1c request, ho\u1eb7c ph\u1ea7n l\u1edbn chi ph\u00ed n\u1eb1m \u1edf output. L\u00fac \u0111\u00f3 c\u00f4ng s\u1ee9c c\u1ea5u h\u00ecnh v\u00e0 theo d\u00f5i cache l\u1edbn h\u01a1n kho\u1ea3n ti\u1ebft ki\u1ec7m \u0111\u01b0\u1ee3c. B\u1ea1n c\u00f3 th\u1ec3 b\u1ecf qua c\u01a1 ch\u1ebf n\u00e0y n\u1ebfu:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Ch\u1ec9 g\u1eedi prompt ng\u1eafn.<\/li>\n\n\n<li>M\u1ed7i t\u00e1c v\u1ee5 ch\u1ec9 c\u00f3 m\u1ed9t request.<\/li>\n\n\n<li>Context thay \u0111\u1ed5i ho\u00e0n to\u00e0n gi\u1eefa c\u00e1c request.<\/li>\n\n\n<li>D\u00f9ng model c\u00f3 input r\u1ea5t r\u1ebb.<\/li>\n\n\n<li>Ph\u1ea7n l\u1edbn chi ph\u00ed \u0111\u1ebfn t\u1eeb output token.<\/li>\n\n\n<li>Workflow ch\u1ee7 y\u1ebfu x\u1eed l\u00fd c\u00e1c nhi\u1ec7m v\u1ee5 \u0111\u1ed9c l\u1eadp.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Ng\u01b0\u1ee3c l\u1ea1i, prompt caching \u0111\u00e1ng quan t\u00e2m khi Hermes Agent ch\u1ea1y nh\u01b0 tr\u1ee3 l\u00fd li\u00ean t\u1ee5c, coding agent, research agent, workflow nhi\u1ec1u b\u01b0\u1edbc ho\u1eb7c ch\u1ea1y d\u00e0i h\u1ea1n.<\/p>\n\n\n\n<h2 id=\"K\u1ebft_lu\u1eadn\">K\u1ebft lu\u1eadn<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">T\u1ed1i \u01b0u chi ph\u00ed API cho Hermes Agent kh\u00f4ng b\u1eaft bu\u1ed9c ph\u1ea3i chuy\u1ec3n sang model r\u1ebb. V\u1edbi workflow d\u00f9ng context l\u1edbn v\u00e0 l\u1eb7p l\u1ea1i nhi\u1ec1u l\u1ea7n, OpenRouter k\u1ebft h\u1ee3p prompt caching gi\u1ea3m r\u00f5 chi ph\u00ed input m\u00e0 v\u1eabn gi\u1eef model \u0111ang d\u00f9ng. \u0110o tr\u01b0\u1edbc b\u1eb1ng <code>hermes prompt-size<\/code>, b\u1eadt cache, r\u1ed3i ki\u1ec3m <code>cached_tokens<\/code> tr\u00ean OpenRouter sau v\u00e0i ng\u00e0y.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">Mu\u1ed1n ch\u1ea1y Hermes Agent 24\/7 m\u00e0 kh\u00f4ng t\u1ef1 c\u00e0i? <a href=\"https:\/\/tino.vn\/vps-hermes\">VPS Hermes Agent c\u1ee7a Tino<\/a> c\u00e0i s\u1eb5n Hermes Agent tr\u00ean VPS NVMe \u0111\u1eb7t t\u1ea1i Vi\u1ec7t Nam; b\u1ea1n v\u1eabn c\u00f3 quy\u1ec1n root \u0111\u1ec3 c\u1ea5u h\u00ecnh. Xem c\u1ea5u h\u00ecnh v\u00e0 gi\u00e1 tr\u00ean trang s\u1ea3n ph\u1ea9m.<\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>B\u00e0i vi\u1ebft li\u00ean quan trong c\u1ee5m Model &#038; chi ph\u00ed Hermes:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/tino.vn\/blog\/nen-chon-llm-nao-cho-hermes-agent\/\">N\u00ean ch\u1ecdn LLM n\u00e0o cho Hermes Agent<\/a><\/li>\n\n\n<li><a href=\"https:\/\/tino.vn\/blog\/llm-tiet-kiem-chi-phi-cho-hermes-agent\/\">LLM gi\u00e1 r\u1ebb cho Hermes Agent<\/a><\/li>\n\n\n<li><a href=\"https:\/\/tino.vn\/blog\/hermes-agent-1-thang-ton-bao-nhieu-chi-phi\/\">Chi ph\u00ed ch\u1ea1y Hermes Agent m\u1ed7i th\u00e1ng<\/a><\/li>\n\n\n<li><a href=\"https:\/\/tino.vn\/blog\/ket-noi-hermes-agent-voi-openrouter-dung-gpt-va-claude\/\">K\u1ebft n\u1ed1i Hermes Agent v\u1edbi OpenRouter<\/a><\/li>\n\n\n<li><a href=\"https:\/\/tino.vn\/blog\/doi-model-llm-cho-hermes-agent\/\">\u0110\u1ed5i model LLM cho Hermes Agent<\/a><\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">B\u00e0i kh\u00e1c v\u1ec1 Hermes Agent: <a href=\"https:\/\/tino.vn\/blog\/hermes-agent-desktop\/\">Hermes Agent Desktop<\/a>, <a href=\"https:\/\/tino.vn\/blog\/ket-noi-google-workspace-voi-hermes-agent\/\">k\u1ebft n\u1ed1i Google Workspace v\u1edbi Hermes Agent<\/a>, <a href=\"https:\/\/tino.vn\/blog\/cach-viet-skill-cho-hermes-agent\/\">c\u00e1ch vi\u1ebft Skill cho Hermes Agent<\/a>, <a href=\"https:\/\/tino.vn\/blog\/hermes-agent-nen-chay-local-hay-vps\/\">Hermes Agent n\u00ean ch\u1ea1y local hay VPS<\/a>.<\/p>\n\n\n\n<h2 id=\"Nh\u1eefng_c\u00e2u_h\u1ecfi_th\u01b0\u1eddng_g\u1eb7p\">Nh\u1eefng c\u00e2u h\u1ecfi th\u01b0\u1eddng g\u1eb7p<\/h2>\n\n\n\t\t<section\t\thelp class=\"sc_fs_faq sc_card    \"\n\t\t\t\t>\n\t\t\t\t<h2 id=\"Hermes_Agent_c\u00f3_t\u1ef1_b\u1eadt_Prompt_Caching_kh\u00f4ng?\">Hermes Agent c\u00f3 t\u1ef1 b\u1eadt Prompt Caching kh\u00f4ng?<\/h2>\t\t\t\t<div>\n\t\t\t\t\t\t<div class=\"sc_fs_faq__content\">\n\t\t\t\t\n\n<p class=\"wp-block-paragraph\">C\u00f3. V\u1edbi Claude qua OpenRouter ho\u1eb7c Anthropic API, Hermes t\u1ef1 \u0111\u1eb7t \u0111i\u1ec3m cache theo <code>prompt_caching.cache_ttl<\/code> (m\u1eb7c \u0111\u1ecbnh 5 ph\u00fat trong file c\u1ea5u h\u00ecnh m\u1eabu). C\u00e1c nh\u00e0 cung c\u1ea5p t\u1ef1 cache nh\u01b0 OpenAI, DeepSeek, Gemini th\u00ec kh\u00f4ng c\u1ea7n c\u1ea5u h\u00ecnh th\u00eam.<\/p>\n\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section\t\thelp class=\"sc_fs_faq sc_card    \"\n\t\t\t\t>\n\t\t\t\t<h2 id=\"C\u1eaft_b\u1edbt_c\u00f4ng_c\u1ee5_v\u00e0_k\u1ef9_n\u0103ng_c\u1ee7a_Hermes_c\u00f3_l\u00e0m_Agent_k\u00e9m_th\u00f4ng_minh_\u0111i_kh\u00f4ng?\">C\u1eaft b\u1edbt c\u00f4ng c\u1ee5 v\u00e0 k\u1ef9 n\u0103ng c\u1ee7a Hermes c\u00f3 l\u00e0m Agent k\u00e9m th\u00f4ng minh \u0111i kh\u00f4ng?<\/h2>\t\t\t\t<div>\n\t\t\t\t\t\t<div class=\"sc_fs_faq__content\">\n\t\t\t\t\n\n<p class=\"wp-block-paragraph\">Ch\u1ec9 khi b\u1ea1n t\u1eaft nh\u1ea7m th\u1ee9 \u0111ang c\u1ea7n. T\u1eaft c\u00e1c nh\u00f3m c\u00f4ng c\u1ee5 kh\u00f4ng d\u00f9ng b\u1eb1ng <code>hermes tools<\/code>, gi\u1eef c\u00f4ng c\u1ee5 c\u1ed1t l\u00f5i v\u00e0 t\u1ec7p ghi nh\u1edb d\u1ef1 \u00e1n. C\u1ea7n t\u00ednh n\u0103ng n\u00e0o th\u00ec b\u1eadt l\u1ea1i sau.<\/p>\n\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section\t\thelp class=\"sc_fs_faq sc_card    \"\n\t\t\t\t>\n\t\t\t\t<h2 id=\"Prompt_Caching_c\u00f3_l\u00e0m_c\u00e2u_tr\u1ea3_l\u1eddi_c\u1ee7a_Hermes_Agent_c\u0169_\u0111i_kh\u00f4ng?\">Prompt Caching c\u00f3 l\u00e0m c\u00e2u tr\u1ea3 l\u1eddi c\u1ee7a Hermes Agent c\u0169 \u0111i kh\u00f4ng?<\/h2>\t\t\t\t<div>\n\t\t\t\t\t\t<div class=\"sc_fs_faq__content\">\n\t\t\t\t\n\n<p class=\"wp-block-paragraph\">Kh\u00f4ng. Prompt caching ch\u1ec9 d\u00f9ng l\u1ea1i ph\u1ea7n t\u00ednh to\u00e1n cho context \u0111\u00e3 c\u00f3; model v\u1eabn x\u1eed l\u00fd th\u00f4ng tin m\u1edbi v\u00e0 t\u1ea1o ph\u1ea3n h\u1ed3i m\u1edbi cho t\u1eebng request. Ch\u1ec9 response caching m\u1edbi tr\u1ea3 l\u1ea1i c\u00e2u tr\u1ea3 l\u1eddi c\u0169 khi request gi\u1ed1ng h\u1ec7t.<\/p>\n\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section\t\thelp class=\"sc_fs_faq sc_card    \"\n\t\t\t\t>\n\t\t\t\t<h2 id=\"D\u00f9ng_OpenRouter_c\u00f3_\u0111\u1eaft_h\u01a1n_g\u1ecdi_th\u1eb3ng_API_c\u1ee7a_h\u00e3ng_kh\u00f4ng?\">D\u00f9ng OpenRouter c\u00f3 \u0111\u1eaft h\u01a1n g\u1ecdi th\u1eb3ng API c\u1ee7a h\u00e3ng kh\u00f4ng?<\/h2>\t\t\t\t<div>\n\t\t\t\t\t\t<div class=\"sc_fs_faq__content\">\n\t\t\t\t\n\n<p class=\"wp-block-paragraph\">OpenRouter ghi kh\u00f4ng c\u1ed9ng th\u00eam ph\u00ed v\u00e0o gi\u00e1 suy lu\u1eadn c\u1ee7a nh\u00e0 cung c\u1ea5p, nh\u01b0ng c\u00f3 ph\u00ed khi n\u1ea1p credit. B\u00f9 l\u1ea1i b\u1ea1n c\u00f3 m\u1ed9t t\u00e0i kho\u1ea3n, \u0111\u1ed5i model nhanh v\u00e0 theo d\u00f5i chi ph\u00ed t\u1eadp trung. Xem th\u00eam <a href=\"https:\/\/tino.vn\/blog\/openrouter-la-gi\/\">OpenRouter l\u00e0 g\u00ec<\/a>.<\/p>\n\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t<\/section>\n\t\t\n<script type=\"application\/ld+json\">\n\t{\n\t\t\"@context\": \"https:\/\/schema.org\",\n\t\t\"@type\": \"FAQPage\",\n\t\t\"mainEntity\": [\n\t\t\t\t\t{\n\t\t\t\t\"@type\": \"Question\",\n\t\t\t\t\"name\": \"Hermes Agent c\u00f3 t\u1ef1 b\u1eadt Prompt Caching kh\u00f4ng?\",\n\t\t\t\t\"acceptedAnswer\": {\n\t\t\t\t\t\"@type\": \"Answer\",\n\t\t\t\t\t\"text\": \"<p>C\u00f3. V\u1edbi Claude qua OpenRouter ho\u1eb7c Anthropic API, Hermes t\u1ef1 \u0111\u1eb7t \u0111i\u1ec3m cache theo prompt_caching.cache_ttl (m\u1eb7c \u0111\u1ecbnh 5 ph\u00fat trong file c\u1ea5u h\u00ecnh m\u1eabu). C\u00e1c nh\u00e0 cung c\u1ea5p t\u1ef1 cache nh\u01b0 OpenAI, DeepSeek, Gemini th\u00ec kh\u00f4ng c\u1ea7n c\u1ea5u h\u00ecnh th\u00eam.<\/p>\"\n\t\t\t\t\t\t\t\t\t}\n\t\t\t}\n\t\t\t,\t\t\t\t{\n\t\t\t\t\"@type\": \"Question\",\n\t\t\t\t\"name\": \"C\u1eaft b\u1edbt c\u00f4ng c\u1ee5 v\u00e0 k\u1ef9 n\u0103ng c\u1ee7a Hermes c\u00f3 l\u00e0m Agent k\u00e9m th\u00f4ng minh \u0111i kh\u00f4ng?\",\n\t\t\t\t\"acceptedAnswer\": {\n\t\t\t\t\t\"@type\": \"Answer\",\n\t\t\t\t\t\"text\": \"<p>Ch\u1ec9 khi b\u1ea1n t\u1eaft nh\u1ea7m th\u1ee9 \u0111ang c\u1ea7n. T\u1eaft c\u00e1c nh\u00f3m c\u00f4ng c\u1ee5 kh\u00f4ng d\u00f9ng b\u1eb1ng hermes tools, gi\u1eef c\u00f4ng c\u1ee5 c\u1ed1t l\u00f5i v\u00e0 t\u1ec7p ghi nh\u1edb d\u1ef1 \u00e1n. C\u1ea7n t\u00ednh n\u0103ng n\u00e0o th\u00ec b\u1eadt l\u1ea1i sau.<\/p>\"\n\t\t\t\t\t\t\t\t\t}\n\t\t\t}\n\t\t\t,\t\t\t\t{\n\t\t\t\t\"@type\": \"Question\",\n\t\t\t\t\"name\": \"Prompt Caching c\u00f3 l\u00e0m c\u00e2u tr\u1ea3 l\u1eddi c\u1ee7a Hermes Agent c\u0169 \u0111i kh\u00f4ng?\",\n\t\t\t\t\"acceptedAnswer\": {\n\t\t\t\t\t\"@type\": \"Answer\",\n\t\t\t\t\t\"text\": \"<p>Kh\u00f4ng. Prompt caching ch\u1ec9 d\u00f9ng l\u1ea1i ph\u1ea7n t\u00ednh to\u00e1n cho context \u0111\u00e3 c\u00f3; model v\u1eabn x\u1eed l\u00fd th\u00f4ng tin m\u1edbi v\u00e0 t\u1ea1o ph\u1ea3n h\u1ed3i m\u1edbi cho t\u1eebng request. Ch\u1ec9 response caching m\u1edbi tr\u1ea3 l\u1ea1i c\u00e2u tr\u1ea3 l\u1eddi c\u0169 khi request gi\u1ed1ng h\u1ec7t.<\/p>\"\n\t\t\t\t\t\t\t\t\t}\n\t\t\t}\n\t\t\t,\t\t\t\t{\n\t\t\t\t\"@type\": \"Question\",\n\t\t\t\t\"name\": \"D\u00f9ng OpenRouter c\u00f3 \u0111\u1eaft h\u01a1n g\u1ecdi th\u1eb3ng API c\u1ee7a h\u00e3ng kh\u00f4ng?\",\n\t\t\t\t\"acceptedAnswer\": {\n\t\t\t\t\t\"@type\": \"Answer\",\n\t\t\t\t\t\"text\": \"<p>OpenRouter ghi kh\u00f4ng c\u1ed9ng th\u00eam ph\u00ed v\u00e0o gi\u00e1 suy lu\u1eadn c\u1ee7a nh\u00e0 cung c\u1ea5p, nh\u01b0ng c\u00f3 ph\u00ed khi n\u1ea1p credit. B\u00f9 l\u1ea1i b\u1ea1n c\u00f3 m\u1ed9t t\u00e0i kho\u1ea3n, \u0111\u1ed5i model nhanh v\u00e0 theo d\u00f5i chi ph\u00ed t\u1eadp trung. Xem th\u00eam <a>OpenRouter l\u00e0 g\u00ec<\/a>.<\/p>\"\n\t\t\t\t\t\t\t\t\t}\n\t\t\t}\n\t\t\t\t\t\t]\n\t}\n<\/script>\n","protected":false},"excerpt":{"rendered":"<p>\u0110\u1ec3 t\u1ed1i \u01b0u chi ph\u00ed API cho Hermes Agent, k\u1ebft h\u1ee3p OpenRouter v\u1edbi prompt caching: ph\u1ea7n prompt l\u1eb7p l\u1ea1i \u1edf m\u1ed7i l\u01b0\u1ee3t (system prompt, m\u00f4 t\u1ea3 c\u00f4ng c\u1ee5, b\u1ed9 nh\u1edb) \u0111\u01b0\u1ee3c \u0111\u1ecdc t\u1eeb cache v\u1edbi gi\u00e1 th\u1ea5p h\u01a1n nhi\u1ec1u so v\u1edbi input th\u01b0\u1eddng, v\u1edbi Claude l\u00e0 0,1 l\u1ea7n gi\u00e1 input. Ho\u00e1 \u0111\u01a1n gi\u1ea3m r\u00f5 khi [&hellip;]<\/p>\n","protected":false},"author":23,"featured_media":130443,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7592],"tags":[7593],"class_list":["post-130430","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-hermes-agent","tag-hermes-agent"],"_links":{"self":[{"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/posts\/130430","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/users\/23"}],"replies":[{"embeddable":true,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/comments?post=130430"}],"version-history":[{"count":6,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/posts\/130430\/revisions"}],"predecessor-version":[{"id":132504,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/posts\/130430\/revisions\/132504"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/media\/130443"}],"wp:attachment":[{"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/media?parent=130430"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/categories?post=130430"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tino.vn\/blog\/wp-json\/wp\/v2\/tags?post=130430"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}