You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Free models fail inside official OpenCode TUI with "free tier can only be used from within OpenCode" (Windows, v2.0.32) #56324
OpenCode's free models suddenly stopped working inside the official OpenCode terminal application on Windows.
Every request immediately returns:
Error: OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building.
I am already using the official OpenCode CLI/TUI, so the error appears incorrect.
The issue originally occurred during a normal development session. It persists across existing sessions, brand-new sessions, different free models, and an isolated standalone instance outside my project.
Troubleshooting already performed
Upgraded OpenCode to v2.0.32.
Restarted OpenCode.
Successfully reauthenticated with opencode auth login opencode.
Confirmed OpenCode Console credentials were stored using opencode auth list.
Created a brand-new session outside my project directory.
Launched with opencode --standalone.
Selected the built-in Build agent.
Tested DeepSeek V4.1 Flash Free and Nemotron 3.5 Lightning Free. Both return the identical error.
Checked the resolved configuration using opencode debug config:
Configuration parses successfully.
V2 permissions configuration is present.
No global or Build-agent deny rules detected by the checks performed.
No legacy Zen provider override detected.
No OPENCODE_PERMISSION environment override detected.
Confirmed no MCP servers are configured.
Reauthentication, upgrading, switching models, and starting fresh sessions did not resolve the issue.
Additional context
The free models were working normally before this issue appeared. I was able to complete development tasks using DeepSeek V4.1 Flash Free earlier.
Plugins
No MCP servers are currently configured (opencode mcp list returns "No MCP servers configured"). I previously evaluated code-review-graph but fully uninstalled the integration before this issue. Other global OpenCode plugins have not been verified.
OpenCode version
2.0.32
Steps to reproduce
On Windows, install or upgrade the official OpenCode CLI to v2.0.32.
Authenticate successfully using opencode auth login opencode.
Open PowerShell and run:
cd $HOME
opencode --standalone
In the OpenCode TUI, select the built-in Build agent.
Select DeepSeek V4.1 Flash Free.
Submit: Hello. Reply with one sentence. Do not modify any files.
Observe the error: OpenCode's free tier can only be used from within OpenCode.
The same issue also occurs in ordinary OpenCode sessions without --standalone.
Screenshot and/or share link
No response
Operating System
Windows 11
Terminal
PowerShell on Windows
Reacted by imknown, Ethan Zhang, José Costa, nedaddev1355, bitzlf, Haiu, Raymel, Doonv, AmirArani, Mouayad Alhamwi and 21 more
Thanks for the detailed report and for all the troubleshooting you've already done!
This error comes from the Zen gateway, and it has several known causes that have nothing to do with where you run OpenCode. You've already ruled out a missing Console login. Some of the other causes are easy to miss in a config check, so could you share the following?
The full output of opencode debug config, with any API keys or tokens removed. In particular, check whether read or shell/bash is denied anywhere in any form ("deny", {"*": "deny"}, an agent-level permission, or an old-style "tools": { "bash": false }). Also check whether there's a provider entry pointing at https://opencode.ai/zen/v1. Each of these is known to cause exactly this error.
Your plugins: any plugin entries in your global config (%USERPROFILE%\.config\opencode\opencode.json / opencode.jsonc), and any files in %USERPROFILE%\.config\opencode\plugin(s)\. Since you recently uninstalled code-review-graph, please check that nothing from it is left over.
A log from one failing request: run opencode --print-logs --standalone, send one prompt, and paste the lines around the error, again with any tokens removed.
You can reply here with the details, or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know when you're done.
Related issues, ordered from most to least likely to cover your problem:
Thanks for the detailed report and for all the troubleshooting you've already done!
This error comes from the Zen gateway, and it has several known causes that have nothing to do with where you run OpenCode. You've already ruled out a missing Console login. Some of the other causes are easy to miss in a config check, so could you share the following?
The full output of opencode debug config, with any API keys or tokens removed. In particular, check whether read or shell/bash is denied anywhere in any form ("deny", {"*": "deny"}, an agent-level permission, or an old-style "tools": { "bash": false }). Also check whether there's a provider entry pointing at https://opencode.ai/zen/v1. Each of these is known to cause exactly this error.
Your plugins: any plugin entries in your global config (%USERPROFILE%\.config\opencode\opencode.json / opencode.jsonc), and any files in %USERPROFILE%\.config\opencode\plugin(s)\. Since you recently uninstalled code-review-graph, please check that nothing from it is left over.
A log from one failing request: run opencode --print-logs --standalone, send one prompt, and paste the lines around the error, again with any tokens removed.
You can reply here with the details, or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know when you're done.
Related issues, ordered from most to least likely to cover your problem:
Thanks for following up so fast. I've now captured a failing request using OpenCode v2.0.32 on Windows.
Configuration
The configuration documents I inspected contain:
An OpenCode provider override defining the DeepSeek V4.1 Flash Free model's context limit.
Two local Qwen providers using 127.0.0.1 endpoints.
No visible opencode.ai/zen/v1 provider override.
No visible legacy tools restrictions or explicit read, shell, or bash deny rules.
The resolved configuration parsed successfully. My diagnostic checks detected zero global and Build-agent deny rules.
Plugins
The debug log reports 12 initialized plugins, all with built-in opencode.* names. I found no evidence of code-review-graph loading during this request.
Failing request log
The session starts successfully, but the model request fails almost immediately:
level=ERROR
message="Failed to drain Session"
cause="AI.Error: OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building."
[cause]: AI.Error.Authentication:
OpenCode's free tier can only be used from within OpenCode.
The client subsequently reports session.step.failed and session.execution.failed.
Additional verification
OpenCode Console login succeeded.
Both DeepSeek V4.1 Flash Free and Nemotron 3.5 Lightning Free fail identically.
The issue occurs with the built-in Build agent in a new standalone session outside my project.
Updating OpenCode and reauthenticating did not resolve it.
The logs appear to show a successful official CLI/server startup followed by an authentication-classified error during model inference.
Could you advise what additional diagnostics would distinguish an incorrect request/client signature from a Zen gateway authorization issue?
Thanks for the details and the log. I tried to reproduce this on v2.0.32 (Linux) a few minutes ago, against the real Zen gateway, and free models worked:
opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word" replied normally. So did nemotron-3.5-lightning-free and big-pickle.
It also worked after I added a config like yours: a provider.opencode override setting the context limit for DeepSeek V4.1 Flash Free, plus a local provider on 127.0.0.1.
I ran these signed out, because I don't have a Console account to test with. Since signed-out requests work, there are two likely explanations: the gateway had a short problem (another commenter saw the same error signed out around 06:13 UTC), or it's specific to your signed-in Console account.
You asked how to tell a client problem from a gateway or account problem. These two checks should answer that:
Try again now, in case it was temporary.
Try once signed out: run opencode auth logout opencode, start a new session and send one prompt to DeepSeek V4.1 Flash Free. Afterwards, run opencode auth login opencode to sign back in.
If it works signed out but fails signed in, your client is fine and the problem is on your account's side. models: GLM-5.3-Flash Free missing from the model list for a signed-in Zen account #56112 is a similar case: a free model was switched off for one account in the Console. Only the team can check that. Please email help@anoma.ly with the time you saw the error, and keep account details out of this issue.
If it still fails signed out, please reply here with the time (UTC), the full error and your rough region.
Related: #55435 and #55616 (same error on Windows) and #56112 (free model switched off for a signed-in account).
I completed both requested checks, but unfortunately the issue persists.
Results:
Retested while signed in: Still fails with the same error.
Signed out using opencode auth logout opencode: Started a fresh standalone session and tested DeepSeek V4.1 Flash Free. The exact same error occurred.
Signed back in: Successfully reauthenticated through OpenCode Console, restarted OpenCode, and tested again. Still fails.
Environment:
OS: Windows
Terminal: PowerShell
OpenCode: v2.0.32
Region: Southern California, United States
Latest troubleshooting timestamp: 2026-10-17 06:22:13 UTC
Earlier captured failure: 2026-10-17 06:10:01 UTC
Error:
OpenCode's free tier can only be used from within OpenCode.
Download it at https://opencode.ai and keep building.
The error occurs inside the official OpenCode TUI, including when launched with opencode --standalone.
Since this reproduces both signed in and signed out, it doesn't appear to be specific to my Console account.
Given that the Linux tests succeed, could this be related to Windows client identification, request headers, or a gateway validation issue affecting Windows clients?
I'm happy to run further targeted diagnostics if needed!
Reacted by Torsten Kunkel, Nima, Yan Sheng, Qiumix and Dizzy
Same issue here on arch and ubuntu, only happened after upgrading to v2.0.31 before it was fine. Might be coincidencal, don't know of you changed something on the server at the same time.
Btw
opencode-go/longcat-2.5-preview-free -> works fine
opencode/deepseek-v4.1-flash-free -> does not work
opencode/space-bunny-free -> does not work
opencode/big-pickle -> does not work
Thanks for running those checks so quickly. Since it fails for you even signed out, your Console account isn't the cause.
I tried again on v2.0.32 (Linux, signed out, no config) against the real Zen gateway between 07:19 and 07:21 UTC. DeepSeek V4.1 Flash Free, Space Bunny Free, Big Pickle and Nemotron 3.5 Lightning Free all replied normally. So did opencode run --standalone with your exact prompt. My earlier runs at 06:17–06:19 UTC also worked, while yours failed at 06:22 UTC.
The same client version works from here and fails for you, and @SjurWarEagle sees it on Linux too. So it doesn't look Windows-specific. It looks more like the gateway treats some networks, regions or times differently. That check runs on the gateway, not in the client code here, so a maintainer will need to look at the gateway logs. Your timestamps (06:10:01 and 06:22:13 UTC) and region will help with that.
Two more checks would help narrow it down:
Try a different network, for example a phone hotspot, with any VPN, proxy or corporate network turned off. Then run opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word" and note whether it works or fails, and the time in UTC.
If it still fails, run the same command with --print-logs --log-level debug added, and paste the lines around the error here, with any tokens or keys removed.
Related: #55435 and #55616 (same error, cause not confirmed).
Confirming this on Windows, Desktop app v2.0.32 (official opencode-desktop-win-x64.exe installer). A few data points that may help narrow it down:
Genuinely the official client. The Desktop app runs the shared background service (opencode-cli.exe, confirmed via /api/info). No npm/third-party client involved.
Signed in with a valid Console session. Stored credential is type: oauth, server https://opencode.ai/console, orgName: Personal, created 2026-10-17, expires 30 days out (not expired). Re-authenticating with a fresh device-code sign-in did not help, and signing out / --standalone reproduce it the same way.
I did let opencode (with go-sub) fetch the http-requests, that's why the "prompt" is a bit strange.
Request:
POST /inference/openai/v1/chat/completions HTTP/1.1
Authorization: Bearer public
Content-Type: application/json
User-Agent: opencode/latest/2.0.32/cli
b3: <xxx>
traceparent: 00-e58a8b04e76fec58804e7d236e4d753b-c1ac674d88eb28e3-01
x-opencode-client: cli
x-opencode-project: <xxx>
x-opencode-session: <xxx>
x-opencode-session-id: <xxx>
x-session-affinity: <xxx>
x-session-id: ses_<xxx>
Connection: keep-alive
Accept: */*
Host: opencode.ai
Accept-Encoding: gzip, deflate, br, zstd
Content-Length: 2367
{"model":"deepseek-v4.1-flash-free","messages":[{"role":"system","content":"You are a title generator. You output ONLY a thread title. Nothing else.\n\n<task>\nGenerate a brief title that would
help the user find this conversation later.\n\nFollow all rules in <rules>\nUse the <examples> so you know what a good title looks like.\nYour output must be:\n- A single line\n- <=50 characters
\n- No explanations\n</task>\n\n<rules>\n- you MUST use the same language as the user message you are summarizing\n- Title must be grammatically correct and read naturally - no word salad\n- Nev
er include tool names in the title (e.g. \"read tool\", \"bash tool\", \"edit tool\")\n- Focus on the main topic or question the user needs to retrieve\n- Vary your phrasing - avoid repetitive p
atterns like always starting with \"Analyzing\"\n- When a file is mentioned, focus on WHAT the user wants to do WITH the file, not just that they shared it\n- Keep exact: technical terms, number
s, filenames, HTTP codes\n- Remove: the, this, my, a, an\n- Never assume tech stack\n- Never use tools\n- NEVER respond to questions, just generate a title for the conversation\n- The title shou
ld NEVER include \"summarizing\" or \"generating\" when generating a title\n- DO NOT SAY YOU CANNOT GENERATE A TITLE OR COMPLAIN ABOUT THE INPUT\n- Always output something meaningful, even if th
e input is minimal.\n- If the user message is short or conversational (e.g. \"hello\", \"lol\", \"what's up\", \"hey\"):\n -> create a title that reflects the user's tone or intent (such as Gre
eting, Quick check-in, Light chat, Intro message, etc.)\n</rules>\n\n<examples>\n\"debug 500 errors in production\" -> Debugging production 500 errors\n\"refactor user service\" -> Refactoring u
ser service\n\"why is app.js failing\" -> app.js failure investigation\n\"implement rate limiting\" -> Rate limiting implementation\n\"how do I connect postgres to my API\" -> Postgres API conne
ction\n\"best practices for React hooks\" -> React hooks best practices\n\"@src/credential.ts can you add refresh token support\" -> Credential refresh token support\n\"@utils/parser.ts this is
broken\" -> Parser bug fix\n\"look at @config.json\" -> Config review\n\"@App.tsx add dark mode toggle\" -> Dark mode toggle in App\n</examples>"},{"role":"user","content":"hi"}],"stream":true,"
stream_options":{"include_usage":true}}
{"type":"error","error":{"type":"FreeTierError","message":"OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building."}}
Fresh repro on Windows, Desktop app v2.0.32 (official installer), signed-in Console session — still failing, adding one more confirmed UTC timestamp for gateway-log correlation.
Command:
opencode run --standalone --print-logs --log-level debug --model "opencode/deepseek-v4.1-flash-free" "hi"
Result — fails immediately, 2026-10-17T08:17:24.812Z (UTC):
timestamp=2026-10-17T08:17:24.812Z level=ERROR run=<redacted> message="Failed to drain Session"
cause="AI.Error: OpenCode's free tier can only be used from within OpenCode. Download it at
https://opencode.ai and keep building. ... [cause]: AI.Error.Authentication: OpenCode's free tier
can only be used from within OpenCode."
Error: OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building.
Notes:
The --print-logs output surfaces the client-side error but does not dump the raw gateway HTTP request — @SjurWarEagle's proxy capture above (the actual POST /inference/openai/v1/chat/completions → 403 FreeTierError) is the better source for the wire-level detail. This is just one more confirmed failing timestamp on a different machine/network for the gateway-log lookup.
Confirms the failure on the Desktop app (not just the CLI/TUI), signed-in, Windows, at the timestamp above.
Looks like they replaced the Free tier exceeded usage message with "Error: OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building."
Reacted by Torsten Kunkel, obvirm, Zahra Rafiei, Parnaz Alagheband and Niki
on top of everything, now facing this problem "OpenCode's free models aren't available from your current network because we've detected abusive traffic from this IP address. This isn't related to your account. If you're on a VPN or proxy, turn it off and try again, or subscribe to OpenCode Go at https://opencode.ai/go to keep building."
Still My enveriment is complicated. 1、I run WSL 2、opencode running with a service /usr/local/bin/opencode serve --hostname 0.0.0.0 --port 3322 3、Access web 4、 Input ping get pong then input ping = OpenCode's free models aren't available from your current network because we've detected abusive traffic from this IP address. This isn't related to your account. If you're on a VPN or proxy, turn it off and try again, or subscribe to OpenCode Go at https://opencode.ai/go to keep building.
I'm closing this for now because we couldn't reproduce it, and we're still missing the network test and debug logs from the last comment. On v2.0.32, free models worked for us signed out against the real Zen gateway at 06:17–06:19 and 07:19–07:21 UTC on Oct 17. You saw the error at 06:10 and 06:22 UTC, both signed in and signed out. That points to something on the gateway that depends on your network or region, not on the client.
If it still happens, please reply here and we'll reopen. It would help to include:
whether it also fails on a different network (for example a phone hotspot, with any VPN or proxy turned off)
the time it failed, in UTC
the lines around the error from opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word" --print-logs --log-level debug, with any tokens or keys removed
#55435 and #55616 track the same error, if you'd like to follow them. Thanks for all the troubleshooting you did!
now it's working but after using just for few minutes it's showing "You've used today's free messages. They reset at 6:00 AM · 10/25/2026.
Upgrade to OpenCode Go → /connect · Switch model → /models".
29. This is not a token charity guys stop expecting unlimited free model usage for free , the opencode guys are already very generous. @iusenvimbtw
I don't expect anything from this project or its community. I expect honesty, transparency, and basic respect for users. So what exactly are you talking about?
My issue is simple: if you advertise a service as free, it should actually be free under the conditions users are led to expect. Don't hide limitations behind a Pro subscription, restrict access through paywalls, or burn through users' free tokens and then act as if nothing has changed. If there are limits, say so clearly from the beginning. Free with undisclosed restrictions is not the same as genuinely free.
This kind of behavior wastes people's time, breaks trust, and damages the reputation of the open-source community. Open source should be about transparency and user freedom, not using free access as a promotional hook to funnel people into paid services.
And honestly, why should we keep promoting projects that don't respect the people who use, support, and recommend them?
Respect goes both ways. We need to respect ourselves enough to stop defending misleading practices just because a project is open source or popular. Nobody owes a project blind loyalty, free promotion, or endless patience.
If you change your terms, be honest about it. If your service has limits, disclose them. And if you can no longer provide what you originally advertised, own that decision instead of expecting users to accept it without criticism.
Freesoftware/Open source deserves better than marketing disguised as community goodwill. And users deserve better than broken promises.
Reacted by gdlittlefish, Gennaro Tedesco and Towhidul Islam
Activity
opencode-agent commentedon Oct 17, 2026
Thanks for the detailed report and for all the troubleshooting you've already done!
This error comes from the Zen gateway, and it has several known causes that have nothing to do with where you run OpenCode. You've already ruled out a missing Console login. Some of the other causes are easy to miss in a config check, so could you share the following?
opencode debug config, with any API keys or tokens removed. In particular, check whetherreadorshell/bashis denied anywhere in any form ("deny",{"*": "deny"}, an agent-levelpermission, or an old-style"tools": { "bash": false }). Also check whether there's a provider entry pointing athttps://opencode.ai/zen/v1. Each of these is known to cause exactly this error.pluginentries in your global config (%USERPROFILE%\.config\opencode\opencode.json/opencode.jsonc), and any files in%USERPROFILE%\.config\opencode\plugin(s)\. Since you recently uninstalled code-review-graph, please check that nothing from it is left over.opencode --print-logs --standalone, send one prompt, and paste the lines around the error, again with any tokens removed.You can reply here with the details, or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know when you're done.
Related issues, ordered from most to least likely to cover your problem:
zen/v1provider override or signing in again fixed it.shellorreadpermissions are denied #51241: free models fail with this error whenreadorshellis denied in any form, including"*": "deny".shellis denied on an agent.opencode auth login opencodefixes.PrinOrange commentedon Oct 17, 2026
Same here. I found another resolution of this issue, but I can not ensure whether it's useful. #55737 (comment)
mirabhakta commentedon Oct 17, 2026
Thanks for following up so fast. I've now captured a failing request using OpenCode v2.0.32 on Windows.
Configuration
The configuration documents I inspected contain:
127.0.0.1endpoints.opencode.ai/zen/v1provider override.toolsrestrictions or explicitread,shell, orbashdeny rules.The resolved configuration parsed successfully. My diagnostic checks detected zero global and Build-agent deny rules.
Plugins
The debug log reports 12 initialized plugins, all with built-in
opencode.*names. I found no evidence of code-review-graph loading during this request.Failing request log
The session starts successfully, but the model request fails almost immediately:
The client subsequently reports
session.step.failedandsession.execution.failed.Additional verification
The logs appear to show a successful official CLI/server startup followed by an authentication-classified error during model inference.
Could you advise what additional diagnostics would distinguish an incorrect request/client signature from a Zen gateway authorization issue?
opencode-agent commentedon Oct 17, 2026
Thanks for the details and the log. I tried to reproduce this on v2.0.32 (Linux) a few minutes ago, against the real Zen gateway, and free models worked:
opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word"replied normally. So didnemotron-3.5-lightning-freeandbig-pickle.provider.opencodeoverride setting the contextlimitfor DeepSeek V4.1 Flash Free, plus a local provider on127.0.0.1.I ran these signed out, because I don't have a Console account to test with. Since signed-out requests work, there are two likely explanations: the gateway had a short problem (another commenter saw the same error signed out around 06:13 UTC), or it's specific to your signed-in Console account.
You asked how to tell a client problem from a gateway or account problem. These two checks should answer that:
opencode auth logout opencode, start a new session and send one prompt toDeepSeek V4.1 Flash Free. Afterwards, runopencode auth login opencodeto sign back in.Related: #55435 and #55616 (same error on Windows) and #56112 (free model switched off for a signed-in account).
mirabhakta commentedon Oct 17, 2026
I completed both requested checks, but unfortunately the issue persists.
Results:
opencode auth logout opencode: Started a fresh standalone session and tested DeepSeek V4.1 Flash Free. The exact same error occurred.Environment:
Error:
The error occurs inside the official OpenCode TUI, including when launched with
opencode --standalone.Since this reproduces both signed in and signed out, it doesn't appear to be specific to my Console account.
Given that the Linux tests succeed, could this be related to Windows client identification, request headers, or a gateway validation issue affecting Windows clients?
I'm happy to run further targeted diagnostics if needed!
SjurWarEagle commentedon Oct 17, 2026
Same issue here on arch and ubuntu, only happened after upgrading to v2.0.31 before it was fine. Might be coincidencal, don't know of you changed something on the server at the same time.
Btw
opencode-go/longcat-2.5-preview-free -> works fine
opencode/deepseek-v4.1-flash-free -> does not work
opencode/space-bunny-free -> does not work
opencode/big-pickle -> does not work
opencode-agent commentedon Oct 17, 2026
Thanks for running those checks so quickly. Since it fails for you even signed out, your Console account isn't the cause.
I tried again on v2.0.32 (Linux, signed out, no config) against the real Zen gateway between 07:19 and 07:21 UTC. DeepSeek V4.1 Flash Free, Space Bunny Free, Big Pickle and Nemotron 3.5 Lightning Free all replied normally. So did
opencode run --standalonewith your exact prompt. My earlier runs at 06:17–06:19 UTC also worked, while yours failed at 06:22 UTC.The same client version works from here and fails for you, and @SjurWarEagle sees it on Linux too. So it doesn't look Windows-specific. It looks more like the gateway treats some networks, regions or times differently. That check runs on the gateway, not in the client code here, so a maintainer will need to look at the gateway logs. Your timestamps (06:10:01 and 06:22:13 UTC) and region will help with that.
Two more checks would help narrow it down:
opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word"and note whether it works or fails, and the time in UTC.--print-logs --log-level debugadded, and paste the lines around the error here, with any tokens or keys removed.Related: #55435 and #55616 (same error, cause not confirmed).
SjurWarEagle commentedon Oct 17, 2026
does not work. using mobile hotspot, logged out of opencode go sub
melodiw82 commentedon Oct 17, 2026
Confirming this on Windows, Desktop app v2.0.32 (official
opencode-desktop-win-x64.exeinstaller). A few data points that may help narrow it down:opencode-cli.exe, confirmed via/api/info). No npm/third-party client involved.type: oauth, serverhttps://opencode.ai/console,orgName: Personal, created 2026-10-17,expires30 days out (not expired). Re-authenticating with a fresh device-code sign-in did not help, and signing out /--standalonereproduce it the same way.opencodeprovider entry pointing athttps://opencode.ai/zen/v1anywhere (globalopencode.jsonconly defines my own custom providers + local models). So the rejection isn't coming from a strayzen/v1override.https://opencode.ai/zen/v1/modelsreturns HTTP 200, yet free-tier model requests are rejected — and per Every HTTP 403 is classified as AuthenticationError, so FreeTierError/RegionError reach ACP clients as "Authentication required" #55995 that403 surfaces asAI.Error.Authentication("...can only be used from within OpenCode"). This lines up with Free-tier 403 is network-dependent in official 2.0.32 and 1.18.35 clients #56334's finding that the403 is network/eligibility-dependent at the hosted gateway rather than in the client.Happy to provide any non-sensitive details (version strings, timings, response headers) if useful — just say what would help.
SjurWarEagle commentedon Oct 17, 2026
I did let opencode (with go-sub) fetch the http-requests, that's why the "prompt" is a bit strange.
Parsed body of response:
melodiw82 commentedon Oct 17, 2026
Fresh repro on Windows, Desktop app v2.0.32 (official installer), signed-in Console session — still failing, adding one more confirmed UTC timestamp for gateway-log correlation.
Command:
Result — fails immediately,
2026-10-17T08:17:24.812Z(UTC):Notes:
--print-logsoutput surfaces the client-side error but does not dump the raw gateway HTTP request — @SjurWarEagle's proxy capture above (the actualPOST /inference/openai/v1/chat/completions→403 FreeTierError) is the better source for the wire-level detail. This is just one more confirmed failing timestamp on a different machine/network for the gateway-log lookup.raymelon commentedon Oct 17, 2026
Looks like they replaced the
Free tier exceeded usagemessage with"Error: OpenCode's free tier can only be used from within OpenCode. Download it at https://opencode.ai and keep building."49 remaining items
TowhidMolla-Dev commentedon Oct 18, 2026
on top of everything, now facing this problem "OpenCode's free models aren't available from your current network because we've detected abusive traffic from this IP address. This isn't related to your account. If you're on a VPN or proxy, turn it off and try again, or subscribe to OpenCode Go at https://opencode.ai/go to keep building."
iusenvimbtw commentedon Oct 18, 2026
we can't have nice things in the age of security hell
bryndin commentedon Oct 18, 2026
Works again for me as of today (US). Was broken bad yesterday.
liudasheng commentedon Oct 18, 2026
already work now
TowhidMolla-Dev commentedon Oct 19, 2026
still not working for me (BD), tried many things.
mengxifl commentedon Oct 20, 2026
Still My enveriment is complicated. 1、I run WSL 2、opencode running with a service /usr/local/bin/opencode serve --hostname 0.0.0.0 --port 3322 3、Access web 4、 Input ping get pong then input ping = OpenCode's free models aren't available from your current network because we've detected abusive traffic from this IP address. This isn't related to your account. If you're on a VPN or proxy, turn it off and try again, or subscribe to OpenCode Go at https://opencode.ai/go to keep building.
opencode-agent commentedon Oct 20, 2026
I'm closing this for now because we couldn't reproduce it, and we're still missing the network test and debug logs from the last comment. On v2.0.32, free models worked for us signed out against the real Zen gateway at 06:17–06:19 and 07:19–07:21 UTC on Oct 17. You saw the error at 06:10 and 06:22 UTC, both signed in and signed out. That points to something on the gateway that depends on your network or region, not on the client.
If it still happens, please reply here and we'll reopen. It would help to include:
opencode run -m opencode/deepseek-v4.1-flash-free "Say hello in one word" --print-logs --log-level debug, with any tokens or keys removed#55435 and #55616 track the same error, if you'd like to follow them. Thanks for all the troubleshooting you did!
TowhidMolla-Dev commentedon Oct 24, 2026
now it's working but after using just for few minutes it's showing "You've used today's free messages. They reset at 6:00 AM · 10/25/2026.
Upgrade to OpenCode Go → /connect · Switch model → /models".
esmaeelE commentedon Oct 26, 2026
@TowhidMolla-Dev and everyone I think we must ditch opencode and migrate to another alternatives.
There is no excuse to behave like this.
TowhidMolla-Dev commentedon Oct 27, 2026
is there any free alternative of it you know?
iusenvimbtw commentedon Oct 27, 2026
This is not a token charity guys stop expecting unlimited free model usage for free , the opencode guys are already very generous.
esmaeelE commentedon Oct 27, 2026
I find these two one. Qoder, alphacode. Not exact alternative but good.
esmaeelE commentedon Oct 27, 2026
I don't expect anything from this project or its community. I expect honesty, transparency, and basic respect for users. So what exactly are you talking about?
My issue is simple: if you advertise a service as free, it should actually be free under the conditions users are led to expect. Don't hide limitations behind a Pro subscription, restrict access through paywalls, or burn through users' free tokens and then act as if nothing has changed. If there are limits, say so clearly from the beginning. Free with undisclosed restrictions is not the same as genuinely free.
This kind of behavior wastes people's time, breaks trust, and damages the reputation of the open-source community. Open source should be about transparency and user freedom, not using free access as a promotional hook to funnel people into paid services.
And honestly, why should we keep promoting projects that don't respect the people who use, support, and recommend them?
Respect goes both ways. We need to respect ourselves enough to stop defending misleading practices just because a project is open source or popular. Nobody owes a project blind loyalty, free promotion, or endless patience.
If you change your terms, be honest about it. If your service has limits, disclose them. And if you can no longer provide what you originally advertised, own that decision instead of expecting users to accept it without criticism.
Freesoftware/Open source deserves better than marketing disguised as community goodwill. And users deserve better than broken promises.