[{"data":1,"prerenderedAt":1187},["ShallowReactive",2],{"blog-post-opencode-go-vs-openrouter-vs-deepseek-api":3,"related-posts-opencode-go-vs-openrouter-vs-deepseek-api":1163},{"id":4,"title":5,"author":6,"body":11,"category":1136,"date":1137,"description":1138,"extension":1139,"featured":1140,"hideToc":1140,"image":1141,"imageAlt":1142,"imageHeight":1143,"imageWidth":1144,"itemList":1145,"lang":1145,"meta":1146,"navigation":1147,"noindex":1140,"path":1148,"readingTime":1149,"redirected":1140,"relatedSlugs":1145,"seo":1150,"seoTitle":1151,"stem":1152,"subtitle":1145,"tags":1153,"updatedDate":1137,"__hash__":1162},"blog/blog/opencode-go-vs-openrouter-vs-deepseek-api.md","OpenCode Go vs OpenRouter vs DeepSeek API: Which $10 Lasts Longest for Agents",{"name":7,"role":8,"avatar":9,"bio":10},"Shabnam Katoch","Growth Head","/img/avatars/shabnam-profile.jpeg","Shabnam leads growth at BetterClaw and has spent the past three years building organic distribution for AI-native products. She writes about no-code agent builders, content strategy, and what it actually takes to grow without an ads budget.",{"type":12,"value":13,"toc":1108},"minimark",[14,18,30,98,101,107,110,115,118,241,248,252,261,264,286,293,296,377,380,383,391,397,400,405,412,419,426,519,522,526,529,544,547,594,597,622,695,698,704,707,728,732,735,738,741,744,747,754,757,813,824,828,835,848,851,856,860,875,878,884,890,899,902,905,913,917,920,923,930,934,937,940,947,951,954,957,960,982,985,990,994,1000,1004,1007,1011,1023,1027,1030,1034,1037,1040,1043,1058,1062,1066,1069,1073,1076,1080,1083,1087,1090,1094,1097,1101,1104],[15,16,17],"p",{},"Three ways to pay for the same models. Two defined workloads on each, worked out from list prices on 10 November 2026.",[19,20,22],"callout",{"type":21},"quick-answer",[15,23,24,25,29],{},"In ",[26,27,28],"strong",{},"OpenCode Go vs OpenRouter vs the DeepSeek API",", the difference is how you pay. OpenCode Go is a flat $10 a month with a dollar allowance per model, OpenRouter is pay per token plus a 5.5% credit fee, and DeepSeek charges per token at the source. For DeepSeek V4.1 Flash, Go's $10 covers up to $100 of usage.",[31,32,33,52],"table",{},[34,35,36],"thead",{},[37,38,39,43,46,49],"tr",{},[40,41,42],"th",{},"Quick answer",[40,44,45],{},"OpenCode Go",[40,47,48],{},"OpenRouter",[40,50,51],{},"DeepSeek API",[53,54,55,70,84],"tbody",{},[37,56,57,61,64,67],{},[58,59,60],"td",{},"How you pay",[58,62,63],{},"$10 a month",[58,65,66],{},"Credits plus a 5.5% card fee",[58,68,69],{},"Prepaid balance",[37,71,72,75,78,81],{},[58,73,74],{},"$10 on V4.1 Flash buys",[58,76,77],{},"Up to $100 of usage a month",[58,79,80],{},"$9.20 of usage",[58,82,83],{},"$10 of usage",[37,85,86,89,92,95],{},[58,87,88],{},"Pick it for",[58,90,91],{},"Coding agents",[58,93,94],{},"Many models, one key",[58,96,97],{},"Always-on agents on DeepSeek",[15,99,100],{},"Picture this: it's Sunday night and your coding agent has chewed through a refactor all weekend. One tab says you've used 48% of your weekly Go allowance. Another shows a DeepSeek balance that has barely moved, because weekends are off-peak.",[15,102,103],{},[104,105,106],"em",{},"So which one is actually the cheap one?",[15,108,109],{},"All three, depending on the week. Here's how to tell.",[111,112,114],"h2",{"id":113},"opencode-go-vs-openrouter-vs-deepseek-api-at-a-glance","OpenCode Go vs OpenRouter vs DeepSeek API at a glance",[15,116,117],{},"Checked on each vendor's own pricing page and docs, 10 November 2026.",[31,119,120,132],{},[34,121,122],{},[37,123,124,126,128,130],{},[40,125],{},[40,127,45],{},[40,129,48],{},[40,131,51],{},[53,133,134,148,169,185,199,213,227],{},[37,135,136,139,142,145],{},[58,137,138],{},"Price model",[58,140,141],{},"$10 a month; Go Plus $40; Go Teams $20 per member",[58,143,144],{},"List price per token; 5.5% card fee on credits ($0.80 minimum)",[58,146,147],{},"List price per token, prepaid",[37,149,150,153,156,159],{},[58,151,152],{},"Models",[58,154,155],{},"32 curated",[58,157,158],{},"465 in its models API",[58,160,161,165,166],{},[162,163,164],"code",{},"deepseek-flash"," and ",[162,167,168],{},"deepseek-v4-pro",[37,170,171,174,177,180],{},[58,172,173],{},"Limits",[58,175,176],{},"Dollar allowance per model",[58,178,179],{},"Provider limits only",[58,181,182,183],{},"2,500 concurrent on ",[162,184,164],{},[37,186,187,190,193,196],{},[58,188,189],{},"Off-peak",[58,191,192],{},"Counted at DeepSeek's two rates",[58,194,195],{},"On DeepSeek's own endpoint",[58,197,198],{},"Half price",[37,200,201,204,207,210],{},[58,202,203],{},"Data retention",[58,205,206],{},"0 days on most open models",[58,208,209],{},"Per endpoint; prompts not logged by default",[58,211,212],{},"Stored in China, per its policy",[37,214,215,218,221,224],{},[58,216,217],{},"Key in other tools",[58,219,220],{},"Validated coding agents",[58,222,223],{},"Any OpenAI-compatible tool",[58,225,226],{},"OpenAI or Anthropic format tools",[37,228,229,232,235,238],{},[58,230,231],{},"Spend cap",[58,233,234],{},"The allowance",[58,236,237],{},"Credit limit per key",[58,239,240],{},"Your balance",[15,242,243],{},[244,245],"img",{"alt":246,"src":247},"OpenCode Go vs OpenRouter comparison table with the DeepSeek API: price model, models, limits, off-peak, retention, key use, spend cap","/img/blog/opencode-go-vs-openrouter-vs-deepseek-api-compared.jpg",[111,249,251],{"id":250},"what-opencode-go-is-and-whats-actually-in-it","What OpenCode Go is, and what's actually in it",[15,253,254,255,260],{},"OpenCode Go is a subscription from Anomaly, the team behind OpenCode, that bundles access to a curated list of models. It's optional: OpenCode itself runs with any provider. If you're still picking the agent, our ",[256,257,259],"a",{"href":258},"/blog/hermes-vs-pi-vs-opencode","comparison of Hermes, Pi and OpenCode"," covers that part.",[15,262,263],{},"There are three plans:",[265,266,267,274,280],"ul",{},[268,269,270,273],"li",{},[26,271,272],{},"Go",", $10 a month, for a single-member workspace.",[268,275,276,279],{},[26,277,278],{},"Go Plus",", $40 a month, with two to eight times Go's allowance depending on the model.",[268,281,282,285],{},[26,283,284],{},"Go Teams",", $20 per member a month, each with limits $5 higher than Go.",[15,287,288,289,292],{},"The allowance isn't a request count. OpenCode sets a monthly dollar limit per model and counts your tokens at list price: 20% of it in any 5 hours, 50% in any week. Every response reports what it counted in ",[162,290,291],{},"opencode.allowance",".",[15,294,295],{},"Here's the catch: those dollar limits vary a lot.",[31,297,298,311],{},[34,299,300],{},[37,301,302,305,308],{},[40,303,304],{},"Model on the $10 Go plan",[40,306,307],{},"Monthly limit",[40,309,310],{},"Price per 1M tokens, input / output",[53,312,313,324,335,345,356,367],{},[37,314,315,318,321],{},[58,316,317],{},"DeepSeek V4.1 Flash",[58,319,320],{},"$100",[58,322,323],{},"$0.30 / $1.20 peak, $0.15 / $0.60 off-peak",[37,325,326,329,332],{},[58,327,328],{},"GLM-5.3-Flash",[58,330,331],{},"$60",[58,333,334],{},"$0.15 / $0.50",[37,336,337,340,342],{},[58,338,339],{},"MiMo-V2.6-Flash",[58,341,331],{},[58,343,344],{},"$0.14 / $0.28",[37,346,347,350,353],{},[58,348,349],{},"Qwen3.8 Flash",[58,351,352],{},"$30",[58,354,355],{},"$0.15 / $0.47",[37,357,358,361,364],{},[58,359,360],{},"GLM-5.3",[58,362,363],{},"$15",[58,365,366],{},"$1.40 / $4.40",[37,368,369,372,374],{},[58,370,371],{},"Kimi K3",[58,373,363],{},[58,375,376],{},"$3.00 / $15.00",[15,378,379],{},"Where OpenCode has bulk discounts or reserved GPUs, it passes them on as a higher limit. For new or already-discounted models, \"you still get a little more than if you paid the model providers directly.\"",[15,381,382],{},"So Go is a ten-times deal on V4.1 Flash and a one-and-a-half-times deal on Kimi K3. Same $10.",[15,384,385,386,390],{},"Choosing between the two biggest limits? Our ",[256,387,389],{"href":388},"/blog/deepseek-v4-1-flash-vs-glm-5-3-flash","DeepSeek V4.1 Flash vs GLM 5.3 Flash comparison"," compares price per finished task.",[392,393,394],"blockquote",{},[15,395,396],{},"Go's $10 isn't worth $10 or $100. It's worth whatever your main model's limit says.",[15,398,399],{},"One more surprise: Go is sold on open-weight models, but it also carries Claude Haiku 5.5, GPT 6 Luna, Gemini 4 Argon and Grok 4.7.",[401,402,404],"h3",{"id":403},"does-the-go-key-work-outside-opencode-and-in-dsh","Does the Go key work outside OpenCode, and in DSH?",[15,406,407,408,411],{},"Yes, with conditions. OpenCode says Go is for coding agents that send typical coding-agent traffic, their own user agent and a stable session ID in ",[162,409,410],{},"x-opencode-session",". Validated clients: Claude Code, Codex, Hermes (v0.21.6 or later), Pi, ZCode, jcode and Kilo Code CLI.",[15,413,414,415,418],{},"DSH, DeepSeek's own agent (in developer preview, run as ",[162,416,417],{},"dsh","), accepts any OpenAI-compatible endpoint as a custom model provider. But OpenCode lists it under known problematic clients: session information arrives on some model paths and is missing on others. You can plug the key in; OpenCode just doesn't vouch for it yet.",[15,420,421,422,425],{},"For Claude Code, OpenCode's docs give this ",[162,423,424],{},"~/.claude/settings.json",":",[427,428,433],"pre",{"className":429,"code":430,"language":431,"meta":432,"style":432},"language-json shiki shiki-themes github-light","{\n  \"env\": {\n    \"ANTHROPIC_BASE_URL\": \"https://opencode.ai/inference/go/anthropic\",\n    \"ANTHROPIC_AUTH_TOKEN\": \"oc_sk_...\",\n    \"ANTHROPIC_DEFAULT_SONNET_MODEL\": \"minimax-m3\",\n    \"ANTHROPIC_DEFAULT_HAIKU_MODEL\": \"claude-haiku-5-5\"\n  }\n}\n","json","",[162,434,435,444,454,470,483,496,507,513],{"__ignoreMap":432},[436,437,440],"span",{"class":438,"line":439},"line",1,[436,441,443],{"class":442},"sgsFI","{\n",[436,445,447,451],{"class":438,"line":446},2,[436,448,450],{"class":449},"sYu0t","  \"env\"",[436,452,453],{"class":442},": {\n",[436,455,457,460,463,467],{"class":438,"line":456},3,[436,458,459],{"class":449},"    \"ANTHROPIC_BASE_URL\"",[436,461,462],{"class":442},": ",[436,464,466],{"class":465},"sYBdl","\"https://opencode.ai/inference/go/anthropic\"",[436,468,469],{"class":442},",\n",[436,471,473,476,478,481],{"class":438,"line":472},4,[436,474,475],{"class":449},"    \"ANTHROPIC_AUTH_TOKEN\"",[436,477,462],{"class":442},[436,479,480],{"class":465},"\"oc_sk_...\"",[436,482,469],{"class":442},[436,484,486,489,491,494],{"class":438,"line":485},5,[436,487,488],{"class":449},"    \"ANTHROPIC_DEFAULT_SONNET_MODEL\"",[436,490,462],{"class":442},[436,492,493],{"class":465},"\"minimax-m3\"",[436,495,469],{"class":442},[436,497,499,502,504],{"class":438,"line":498},6,[436,500,501],{"class":449},"    \"ANTHROPIC_DEFAULT_HAIKU_MODEL\"",[436,503,462],{"class":442},[436,505,506],{"class":465},"\"claude-haiku-5-5\"\n",[436,508,510],{"class":438,"line":509},7,[436,511,512],{"class":442},"  }\n",[436,514,516],{"class":438,"line":515},8,[436,517,518],{"class":442},"}\n",[15,520,521],{},"Before you reach for DeepSeek through Go, know this: Go blocks DeepSeek V4.1 Flash and V4 Pro by default. An owner or admin has to select Global under Regions in Settings › Privacy.",[111,523,525],{"id":524},"what-10-buys-on-each-with-the-working","What $10 buys on each, with the working",[15,527,528],{},"Two workloads, both on DeepSeek V4.1 Flash, the one model all three sell at the same list price.",[15,530,531,534,535,538,539,543],{},[26,532,533],{},"Workload A"," is OpenCode's own typical V4.1 Flash request from its Go docs: 410 new input tokens, 71,300 cached input tokens and 310 output tokens. That's a coding agent re-reading a warm context. ",[26,536,537],{},"Workload B"," is the agent run from our ",[256,540,542],{"href":541},"/blog/deepseek-v4-1-flash-agents","DeepSeek V4.1 Flash post",": 20,000 new input tokens, nothing cached, 5,000 output tokens.",[15,545,546],{},"DeepSeek's list price per 1M tokens is $0.30 input, $0.006 cached and $1.20 output at peak, half that off-peak. Go and DeepSeek's endpoint on OpenRouter use the same rates.",[31,548,549,564],{},[34,550,551],{},[37,552,553,556,559,562],{},[40,554,555],{},"Cost per request",[40,557,558],{},"Working at peak",[40,560,561],{},"Peak",[40,563,189],{},[53,565,566,580],{},[37,567,568,571,574,577],{},[58,569,570],{},"A, coding turn",[58,572,573],{},"$0.000123 + $0.000428 + $0.000372",[58,575,576],{},"$0.00092",[58,578,579],{},"$0.00046",[37,581,582,585,588,591],{},[58,583,584],{},"B, agent run",[58,586,587],{},"$0.0060 + $0.0060",[58,589,590],{},"$0.0120",[58,592,593],{},"$0.0060",[15,595,596],{},"Then what $10 gets you:",[265,598,599,605,616],{},[268,600,601,604],{},[26,602,603],{},"DeepSeek API:"," $10 of balance, with no purchase fee listed.",[268,606,607,610,611,615],{},[26,608,609],{},"OpenRouter:"," $9.20 of credit, because the 5.5% fee has a $0.80 minimum that applies below about $14.55 (our ",[256,612,614],{"href":613},"/tools/openrouter-fee-calculator","OpenRouter fee calculator"," runs any amount).",[268,617,618,621],{},[26,619,620],{},"OpenCode Go:"," one month, up to $100 of V4.1 Flash at the same prices, capped at $20 per 5 hours and $50 per week.",[31,623,624,643],{},[34,625,626],{},[37,627,628,631,634,637,640],{},[40,629,630],{},"Requests per $10",[40,632,633],{},"A, peak",[40,635,636],{},"A, off-peak",[40,638,639],{},"B, peak",[40,641,642],{},"B, off-peak",[53,644,645,661,678],{},[37,646,647,649,652,655,658],{},[58,648,51],{},[58,650,651],{},"10,837",[58,653,654],{},"21,673",[58,656,657],{},"833",[58,659,660],{},"1,667",[37,662,663,666,669,672,675],{},[58,664,665],{},"OpenRouter, DeepSeek endpoint",[58,667,668],{},"9,970",[58,670,671],{},"19,939",[58,673,674],{},"767",[58,676,677],{},"1,533",[37,679,680,683,686,689,692],{},[58,681,682],{},"OpenCode Go, monthly ceiling",[58,684,685],{},"108,366",[58,687,688],{},"216,732",[58,690,691],{},"8,333",[58,693,694],{},"16,667",[15,696,697],{},"Calculated from list prices on 10 November 2026, not measured. OpenCode's own estimate for workload A, 216,670 a month, matches our off-peak figure.",[15,699,700],{},[244,701],{"alt":702,"src":703},"What $10 buys on OpenCode Go vs OpenRouter vs DeepSeek API: requests per $10 at peak and off-peak, calculated from list prices","/img/blog/opencode-go-vs-openrouter-vs-deepseek-api-what-10-buys.jpg",[15,705,706],{},"So Go wins by ten times? On paper. Three things shrink it:",[708,709,710,716,722],"ol",{},[268,711,712,715],{},[26,713,714],{},"The weekly cap."," Half the month per week means a heavy first week stops at $50, about 54,000 coding turns at peak.",[268,717,718,721],{},[26,719,720],{},"Your traffic type."," Go is built for coding agents, and OpenCode monitors traffic for abuse.",[268,723,724,727],{},[26,725,726],{},"Your volume."," Below roughly 10,800 coding turns a month at peak, DeepSeek direct costs less than the subscription.",[111,729,731],{"id":730},"opencode-go-vs-openrouter-allowance-vs-variety","OpenCode Go vs OpenRouter: allowance vs variety",[15,733,734],{},"OpenRouter is the opposite bet: no allowance, just the provider's list price per token across hundreds of models behind one key.",[15,736,737],{},"The fee is the price of that: 5.5% on card purchases ($0.80 minimum), 5% on crypto, and your own provider keys run fee-free up to $25,000 of list-price usage a month, then 5%.",[15,739,740],{},"Now the twist.",[15,742,743],{},"For V4.1 Flash 1017, OpenRouter listed 51 endpoints on 10 November. DeepSeek's own endpoint charges DeepSeek's rates, off-peak included. Of the others, 36 were priced below DeepSeek's peak rate on both input and output, and 21 of those 36 state fp8 or fp4 quantization.",[15,745,746],{},"So \"Is OpenRouter cheaper than DeepSeek?\" can honestly be yes. DeepInfra's fp8 endpoint listed $0.098 input and $0.294 output per million, which puts workload B at about $0.0034 a run, or 2,682 runs for $9.20. The trade is a quantized copy on a different host.",[15,748,749,750,292],{},"For V4.1 Flash against other low-cost models, see our ranking of the ",[256,751,753],{"href":752},"/blog/cheapest-ai-models-for-agents","cheapest AI models for agents",[15,755,756],{},"Default routing weighs price, speed, reliability and tool-call quality. To pin DeepSeek itself, set the provider order and turn fallbacks off:",[427,758,760],{"className":429,"code":759,"language":431,"meta":432,"style":432},"{\n  \"model\": \"deepseek/deepseek-v4.1-flash-1017\",\n  \"provider\": { \"order\": [\"deepseek\"], \"allow_fallbacks\": false }\n}\n",[162,761,762,766,778,809],{"__ignoreMap":432},[436,763,764],{"class":438,"line":439},[436,765,443],{"class":442},[436,767,768,771,773,776],{"class":438,"line":446},[436,769,770],{"class":449},"  \"model\"",[436,772,462],{"class":442},[436,774,775],{"class":465},"\"deepseek/deepseek-v4.1-flash-1017\"",[436,777,469],{"class":442},[436,779,780,783,786,789,792,795,798,801,803,806],{"class":438,"line":456},[436,781,782],{"class":449},"  \"provider\"",[436,784,785],{"class":442},": { ",[436,787,788],{"class":449},"\"order\"",[436,790,791],{"class":442},": [",[436,793,794],{"class":465},"\"deepseek\"",[436,796,797],{"class":442},"], ",[436,799,800],{"class":449},"\"allow_fallbacks\"",[436,802,462],{"class":442},[436,804,805],{"class":449},"false",[436,807,808],{"class":442}," }\n",[436,810,811],{"class":438,"line":472},[436,812,518],{"class":442},[15,814,815,816,819,820,823],{},"Or append ",[162,817,818],{},":exacto"," to the model ID to sort by tool-calling reliability, or ",[162,821,822],{},":floor"," to sort by price.",[401,825,827],{"id":826},"the-free-router-charges-people-hit","The free router charges people hit",[15,829,830,831,834],{},"OpenRouter's ",[162,832,833],{},":free"," models allow 20 requests a minute and 50 a day, or 1,000 a day once you've bought at least $10 of credits. Failed attempts count toward that daily cap.",[15,836,837,838,840,841,844,845,847],{},"The surprise bill comes from fallbacks. If you list a ",[162,839,833],{}," model in a ",[162,842,843],{},"models"," array alongside paid ones, OpenRouter falls back to the paid model when the free one is busy, and bills it at its own price. To stay at zero, list only ",[162,846,833],{}," models, or use the Free Models Router on its own.",[15,849,850],{},"On spend caps, OpenRouter wins: each key can carry a credit limit that resets daily, weekly or monthly. Go has the allowance; DeepSeek has your balance.",[392,852,853],{},[15,854,855],{},"Pick Go when you know your model. Pick OpenRouter when you don't, or when you want the router to keep you running.",[111,857,859],{"id":858},"opencode-go-vs-deepseek-api-direct","OpenCode Go vs DeepSeek API direct",[15,861,862,863,866,867,869,870,874],{},"Going direct skips the middle layer: base URL ",[162,864,865],{},"https://api.deepseek.com",", model ",[162,868,164],{},", a prepaid balance. Our ",[256,871,873],{"href":872},"/blog/openrouter-vs-direct-api-agents","OpenRouter vs direct API guide for agents"," covers the same trade for other providers.",[15,876,877],{},"Three things make DeepSeek direct cheaper than it looks.",[15,879,880,883],{},[26,881,882],{},"Off-peak first."," Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays; every other hour is half price. Go counts V4.1 Flash at the same two rates, so off-peak doubles a Go allowance too.",[15,885,886,889],{},[26,887,888],{},"Then cache."," A hit costs $0.006 per million against $0.30 for a miss, 98% off. Even so, in workload A the 71,300 cached tokens are the biggest line on the bill, so keep the system prompt and tools first and unchanged.",[15,891,892,895,896,898],{},[26,893,894],{},"And concurrency:"," 2,500 requests at once per account on ",[162,897,164],{},", counted per account rather than per key. OpenCode publishes no concurrency figure for Go.",[15,900,901],{},"Two costs to plan for. Search and Fetch are $0.75 per 1,000 successful requests each, with no off-peak discount, including server-side calls through DSH, Claude Code or Codex.",[15,903,904],{},"Where direct beats Go: always-on agents that aren't coding traffic, volume above Go's weekly cap, and DeepSeek V4 Pro, where Go's monthly limit is only $15.",[15,906,907,908,912],{},"The cost of going direct is that DeepSeek's capacity is your ceiling, with no router to fail over. When it's overloaded, our ",[256,909,911],{"href":910},"/blog/openclaw-deepseek-503-errors","guide to OpenClaw DeepSeek 503 errors"," covers the fixes.",[111,914,916],{"id":915},"using-any-of-them-in-betterclaw","Using any of them in BetterClaw",[15,918,919],{},"If your agent runs a daily job rather than a coding session (inbox triage, lead scoring, reports), these keys still work. BetterClaw runs managed always-on agents on your own keys, with zero markup across 28+ providers and 500+ connectors, and OpenRouter, DeepSeek and OpenCode all sit in its provider list.",[15,921,922],{},"Two honest limits. BetterClaw isn't a coding agent or an editor. And Go is built for coding-agent traffic, so for an always-on agent we'd use a DeepSeek or OpenRouter key rather than a Go subscription.",[15,924,925,926],{},"Plans start from $19 a month, and your first workspace starts with a 7-day free trial. ",[256,927,929],{"href":928},"/pricing","See BetterClaw plans",[111,931,933],{"id":932},"opencode-go-vs-claude-code-codex-and-claude-pro","OpenCode Go vs Claude Code, Codex and Claude Pro",[15,935,936],{},"Claude Pro is $20 a month ($17 billed annually) and includes Claude Code with 5-hour and weekly limits. ChatGPT Plus is $20 a month with Codex, which also comes with ChatGPT Go at $8.",[15,938,939],{},"Those plans buy one vendor's frontier models. Go buys a basket of mostly cheaper ones, and its only Anthropic model is Claude Haiku 5.5.",[15,941,942,943,946],{},"The useful part: they stack. OpenCode validates Go inside Claude Code and Codex, so you can keep a vendor plan for hard problems and point a separate config at a Go key for volume work (OpenCode says to run ",[162,944,945],{},"/logout"," first if you're signed in to Claude).",[111,948,950],{"id":949},"data-retention-where-your-prompts-go","Data retention: where your prompts go",[15,952,953],{},"OpenCode Go publishes a table per model. Most open models show 0 days and no training; Claude Haiku 5.5, the GPT Luna models and Grok 4.7 show 30 days. The Muse Spark Contributor models train on your data for a lower price, so Go blocks them until an admin allows it.",[15,955,956],{},"DeepSeek on Go shows 0 days under a zero-retention agreement OpenCode renews monthly. The current one runs through 30 November 2026.",[15,958,959],{},"GLM-5.3-Flash comes in two forms. The paid Go entry shows 0 days and no training, and GLM-5.3-Flash Free is free on OpenCode for a limited time from a zero-retention provider that doesn't train on your data. That free one only works inside OpenCode: other tools get HTTP 403, even with a Console key.",[15,961,962,963,966,967,970,971,966,974,977,978,981],{},"OpenRouter logs request metadata but not prompts or completions by default. You can set ",[162,964,965],{},"data_collection"," to ",[162,968,969],{},"deny"," or ",[162,972,973],{},"zdr",[162,975,976],{},"true"," per request. One catch: DeepSeek's own endpoint isn't on OpenRouter's zero-retention list, so ",[162,979,980],{},"zdr: true"," sends V4.1 Flash to third-party hosts.",[15,983,984],{},"DeepSeek direct follows its privacy policy, updated 10 February 2026: personal data is stored in China, and user input is among the data used to train its models, with a right to opt out.",[392,986,987],{},[15,988,989],{},"If zero retention matters, the route you pick matters more than the model you pick.",[111,991,993],{"id":992},"which-to-pick-solo-coder-always-on-agent-team","Which to pick: solo coder, always-on agent, team",[15,995,996],{},[244,997],{"alt":998,"src":999},"Decision flow for OpenCode Go vs OpenRouter vs DeepSeek API by reader type: solo coder, always-on agent, team","/img/blog/opencode-go-vs-openrouter-vs-deepseek-api-decision.jpg",[401,1001,1003],{"id":1002},"solo-coder-in-opencode-claude-code-or-codex","Solo coder in OpenCode, Claude Code or Codex",[15,1005,1006],{},"Pick Go, if V4.1 Flash or GLM-5.3-Flash is your main model and you'd spend more than $10 a month at list. Turn on the Global region for DeepSeek, schedule long runs for off-peak, and keep an eye on the weekly bar.",[401,1008,1010],{"id":1009},"always-on-agent","Always-on agent",[15,1012,1013,1014,1017,1018,1022],{},"Pick DeepSeek direct and run batch jobs off-peak. If you need fallbacks or zero retention, use OpenRouter and pin providers. Going direct, pass ",[162,1015,1016],{},"reasoning_content"," back on tool turns; our notes on ",[256,1019,1021],{"href":1020},"/blog/deepseek-v4-ai-agent-setup-bugs","DeepSeek V4 agent setup bugs"," cover that and two other traps.",[401,1024,1026],{"id":1025},"team","Team",[15,1028,1029],{},"Go Teams gives every member their own limits at $20 each. OpenRouter gives you five workspaces on its Standard plan, each with keys that carry their own credit caps, which is what saves the month for a team that mixes models.",[111,1031,1033],{"id":1032},"the-10-question-is-really-a-traffic-question","The $10 question is really a traffic question",[15,1035,1036],{},"We started with a price question and ended with a traffic one. All three sell V4.1 Flash at the same price per token; what changes is who absorbs your usage.",[15,1038,1039],{},"Go absorbs it for coding agents, up to a cap. OpenRouter charges a small fee to let you go anywhere. DeepSeek charges the least, if you can wait for the clock.",[15,1041,1042],{},"Know what your agent sends, and when. The cheapest $10 follows from that.",[19,1044,1046,1051],{"type":1045},"cta",[15,1047,1048],{},[26,1049,1050],{},"Not sure which bill your agents should run on?",[15,1052,1053,1054],{},"If the model bill is one line among many in your business, start with a conversation instead of a spreadsheet. Our AI automation audit begins with a 30-minute discovery call about how your team works. Within five business days you get a report with a workflow map, a proposed agent architecture, ROI projections and an implementation timeline. ",[256,1055,1057],{"href":1056},"/ai-automation-audit","Book an AI automation audit",[111,1059,1061],{"id":1060},"frequently-asked-questions","Frequently Asked Questions",[401,1063,1065],{"id":1064},"is-opencode-go-worth-it","Is OpenCode Go worth it?",[15,1067,1068],{},"Yes, if you run a coding agent on a model with a generous Go limit and would spend more than $10 a month at list price. DeepSeek V4.1 Flash gets $100 of usage on the $10 plan, but Kimi K3 and GLM-5.3 get only $15.",[401,1070,1072],{"id":1071},"what-models-does-opencode-go-include","What models does OpenCode Go include?",[15,1074,1075],{},"Thirty-two as of 10 November 2026, including DeepSeek V4.1 Flash and V4 Pro, GLM-5.3 and GLM-5.3-Flash, Kimi K3, MiMo, Qwen3.8 and MiniMax, plus Claude Haiku 5.5, GPT 6 Luna, Gemini 4 Argon and Grok 4.7. OpenCode says the list changes as it tests new models.",[401,1077,1079],{"id":1078},"is-openrouter-cheaper-than-the-deepseek-api","Is OpenRouter cheaper than the DeepSeek API?",[15,1081,1082],{},"Not on DeepSeek's own endpoint: OpenRouter passes through DeepSeek's price, then adds a 5.5% fee when you buy credits, with a $0.80 minimum. It can be cheaper if you let it route to third-party hosts, many of which list lower prices for fp8 or fp4 quantized versions of the model.",[401,1084,1086],{"id":1085},"can-i-use-the-opencode-go-key-in-other-tools","Can I use the OpenCode Go key in other tools?",[15,1088,1089],{},"Yes, in coding agents that send coding-agent traffic, their own user agent and a session ID. OpenCode has validated Claude Code, Codex, Hermes, Pi, ZCode, jcode and Kilo Code CLI. It lists DeepSeek's DSH as a known problematic client because session information is missing on some paths.",[401,1091,1093],{"id":1092},"is-glm-53-flash-free-in-opencode","Is GLM-5.3-Flash free in OpenCode?",[15,1095,1096],{},"GLM-5.3-Flash Free is free on OpenCode for a limited time, and its provider follows a zero-retention policy with no training on your data. It only works inside OpenCode, with a daily request limit; other tools get HTTP 403. The paid GLM-5.3-Flash on Go costs $0.15 input and $0.50 output per million tokens, with a $60 monthly limit.",[401,1098,1100],{"id":1099},"how-do-opencode-go-usage-limits-work","How do OpenCode Go usage limits work?",[15,1102,1103],{},"Each model has a monthly limit in dollars, counted at list price. You can use 20% of it in any 5 hours and 50% in any week. If you add Console credits and turn on Extra Usage, requests continue from your balance after the limit instead of stopping.",[1105,1106,1107],"style",{},"html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":432,"searchDepth":446,"depth":446,"links":1109},[1110,1111,1114,1115,1118,1119,1120,1121,1122,1127,1128],{"id":113,"depth":446,"text":114},{"id":250,"depth":446,"text":251,"children":1112},[1113],{"id":403,"depth":456,"text":404},{"id":524,"depth":446,"text":525},{"id":730,"depth":446,"text":731,"children":1116},[1117],{"id":826,"depth":456,"text":827},{"id":858,"depth":446,"text":859},{"id":915,"depth":446,"text":916},{"id":932,"depth":446,"text":933},{"id":949,"depth":446,"text":950},{"id":992,"depth":446,"text":993,"children":1123},[1124,1125,1126],{"id":1002,"depth":456,"text":1003},{"id":1009,"depth":456,"text":1010},{"id":1025,"depth":456,"text":1026},{"id":1032,"depth":446,"text":1033},{"id":1060,"depth":446,"text":1061,"children":1129},[1130,1131,1132,1133,1134,1135],{"id":1064,"depth":456,"text":1065},{"id":1071,"depth":456,"text":1072},{"id":1078,"depth":456,"text":1079},{"id":1085,"depth":456,"text":1086},{"id":1092,"depth":456,"text":1093},{"id":1099,"depth":456,"text":1100},"Comparison","2026-11-12","OpenCode Go vs OpenRouter vs the DeepSeek API for $10: what each gives you, models, limits, data retention and which lasts longest for real agent work.","md",false,"/img/blog/opencode-go-vs-openrouter-vs-deepseek-api.jpg","OpenCode Go vs OpenRouter vs DeepSeek API cover: three price tags labelled flat, per token plus fee and per token",512,1024,null,{},true,"/blog/opencode-go-vs-openrouter-vs-deepseek-api","11 min read",{"title":5,"description":1138},"OpenCode Go vs OpenRouter vs DeepSeek API: Which Lasts","blog/opencode-go-vs-openrouter-vs-deepseek-api",[1154,1155,1156,1157,1158,1159,1160,1161],"opencode go vs openrouter","opencode go vs deepseek api","opencode go","opencode go models","opencode go api key","deepseek api vs openrouter","openrouter free models","glm-5.3-flash free","xUszkei6rCPYkpZlsqyTZqC1ArBgXob9w53n3tsp1WE",[1164,1172,1180],{"stem":1165,"title":1166,"description":1167,"date":1168,"image":1169,"category":1136,"redirected":1140,"author":1170,"readingTime":1171},"blog/qwen-4-vs-qwen-3-8","Qwen 4: Release Date, Expected Sizes and What Changes from Qwen 3.8","Qwen 4 is not out yet. What Qwen has confirmed, what the leaked model ids suggest, the expected sizes including 27B, and what to run until it ships.","2026-11-11","/img/blog/qwen-4-vs-qwen-3-8.jpg",{"name":7,"role":8,"avatar":9,"bio":10},"9 min read",{"stem":1173,"title":1174,"description":1175,"date":1176,"image":1177,"category":1178,"redirected":1140,"author":1179,"readingTime":1171},"blog/is-opus-5-5-nerfed","Is Opus 5.5 Nerfed? How to Check Which Model Your Claude Code Session Actually Ran","Is Opus 5.5 nerfed? How to check which model your Claude Code session used, what the safeguards fallback does and what Anthropic has said.","2026-11-10","/img/blog/is-opus-5-5-nerfed.jpg","Troubleshooting",{"name":7,"role":8,"avatar":9,"bio":10},{"stem":1181,"title":1182,"description":1183,"date":1176,"image":1184,"category":1178,"redirected":1140,"author":1185,"readingTime":1186},"blog/openclaw-fallback-model-not-working","OpenClaw Fallback Model Not Working After 2026.10.10: Why","OpenClaw fallback model not working after 2026.10.10? Scheduled jobs now stay on one provider by design. What still works, the config and 3 workarounds.","/img/blog/openclaw-fallback-model-not-working.jpg",{"name":7,"role":8,"avatar":9,"bio":10},"8 min read",1794489130505]