5.6 Luna on opencode zen tui showing half the price being shown on web version #39891

Activity

1. github-actions

   assigned

   MrMushrooooom

   on Jul 31, 2026
2. tejush1998 commented on Jul 31, 2026

   

   tejush1998

   on Jul 31, 2026

   Last edited by tejush1998

   Author

   More actions

   Logged this in opencode in packages/opencode/src/session/llm.ts :
   Response from openai luna :

   {
     "inputTokens": 82307,
     "inputTokenDetails": {
       "noCacheTokens": 3,
       "cacheReadTokens": 79354,
       "cacheWriteTokens": 2950
     },
   ....
   }
   

   Meanwhile opencode usage shows:

   Total input: 85257 
   Input: 2953
   Cached input: 79394
   Cached write: 2950 (leftover)
   

   As you can see above inputTokens: 82307 vs 85257
   Diff = cache_write_tokens = 2950 (inputTokens from openai is exclusive of cacheWriteTokens)

   Problem?

   the getUsage function in packages/opencode/src/session/session.ts :
   assumes cacheWriteInputTokens to be part of inputTokens

     const adjustedInputTokens = safe(inputTokens - cacheReadInputTokens - cacheWriteInputTokens)
   

   This shows in tui as low billing than it actually is, because cacheWrites are eating into the freshTokens total

   Anthropic works properly:

   {
     "inputTokens": 23335,
     "inputTokenDetails": {
       "noCacheTokens": 2,
       "cacheReadTokens": 23028,
       "cacheWriteTokens": 305
     },
   ...
   }
   

   Opencode usage correctly showing totalInput : 23335, which matches the above inputTokens

   Question: So this is an upstream issue need to be reported to OpenAI? Or should we handle it? Would love to contribute
3. Zaczero commented on Aug 18, 2026

   

   Zaczero

   on Aug 18, 2026

   More actions

   The TUI is not “about half by rounding.” It is subtracting Luna’s cache-write tokens from a total that never contained them as extra, and the web is not doing that subtraction.

   Your own capture is enough:

   OpenAI Luna
     inputTokens:              82,307
       noCacheTokens:               3
       cacheReadTokens:        79,354
       cacheWriteTokens:        2,950
   
     3 + 79,354 + 2,950 = 82,307
   

   So inputTokens is already the whole prompt: fresh + read + write. It is not exclusive of the write.

   TUI then shows:

   Total input:    85,257     = 82,307 + 2,950   (write added again)
   Input:           2,953     = 82,307 − 79,354  (write still inside)
   Cached write:    2,950
   

   getUsage then does the Anthropic split:

   fresh = inputTokens − cacheRead − cacheWrite
         = 82,307 − 79,354 − 2,950
         = 3
   

   Those 2,950 write tokens were already inside 82,307. Subtracting them again leaves 3 tokens at the input rate instead of 2,953. Anthropic looks right because its inputTokens is exclusive of cache, so the same minus does not eat a second copy.

   The web path does not subtract the write. Zen only removes the cache read:

   input billed = prompt_tokens − cached_tokens   → 82,307 − 79,354 = 2,953
   

   plus 2,950 at the write rate. Same tokens, two bills on the web, one bill (write only) on the TUI.

   On this request that is 2,950 tokens at Luna’s input price that the TUI never charges:

   low tier   2,950 × $0.20 / 1M = $0.0006
   high tier  2,950 × $0.40 / 1M = $0.0012
   

   One row is small. A long session is thousands of those writes, which is the “TUI looks like half the web” screenshot.

   This is not an OpenAI bug. 3 + 79,354 + 2,950 = 82,307 is internally consistent. The bug is applying the exclusive split (input − read − write) to a provider whose inputTokens is already inclusive. Handle it on the OpenAI / Luna mapper: if noCache + read + write == inputTokens, take noCache as fresh and do not subtract write a second time.

   opencode/packages/ai/src/protocols/openai-chat.ts

   Lines 582 to 589 in 9e22c40

   	const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
   	if (!usage) return undefined
   	const input = usage.prompt_tokens ?? undefined
   	const output = usage.completion_tokens ?? undefined
   	const cached = usage.prompt_tokens_details?.cached_tokens ?? undefined
   	const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
   	const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
   	const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
4. thdxr commented on Oct 11, 2026

   

   thdxr

   on Oct 11, 2026

   Member

   More actions

   Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
5. thdxr

   closed this as not plannedon Oct 11, 2026