anomalyco / anomalyco/opencode

Incorrect session cost estimate for models with context tiered pricing

Open
#42,910 4 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
TypeScript
Stars
209k
Forks
27.5k
PR merge metrics
PR metrics pending

Description

Description

Opencode's pricing estimate is incorrect when a model with with tiered context pricing and your session goes into the higher tier. The local cost estimates use the price of smaller context for the entire session, resulting in a cost estimate that is lower than what it should be.

For example, the session that I discovered this showed $3.37 in opencode's UI, but the real price was $3.71. The longer the session, the wider the discrepancy.

Testing with grok 4.6, which has different pricing for > 200k, the model calls reported in Zen differed from what is stored in the local sqlite database.

Plugins

No response

OpenCode version

1.18.18

Steps to reproduce

For testing with Zen:
Use a model that has tiered context pricing, such as grok 4.6 or one of the gpt 5.6 models. Have a session go into the higher tiered pricing. Compare local session pricing vs Zen's UI.

   WITH ranked_calls AS (
     SELECT
       ROW_NUMBER() OVER (ORDER BY time_created, id) AS call_number,
       time_created,
       data
     FROM message
     WHERE session_id = 'ses_ff50d39fbffe2mOJcly5gtxQyW'
       AND json_extract(data, '$.role') = 'assistant'
       AND (
         COALESCE(json_extract(data, '$.tokens.input'), 0) +
         COALESCE(json_extract(data, '$.tokens.cache.read'), 0) +
         COALESCE(json_extract(data, '$.tokens.cache.write'), 0) +
         COALESCE(json_extract(data, '$.tokens.output'), 0) +
         COALESCE(json_extract(data, '$.tokens.reasoning'), 0)
       ) > 0
   )
   SELECT
     call_number,
     strftime('%b %d, %H:%M', time_created / 1000, 'unixepoch', 'localtime') AS date,
     json_extract(data, '$.modelID') AS model,
     COALESCE(json_extract(data, '$.tokens.input'), 0) +
       COALESCE(json_extract(data, '$.tokens.cache.read'), 0) +
       COALESCE(json_extract(data, '$.tokens.cache.write'), 0) AS input,
     printf('$%.4f', COALESCE(json_extract(data, '$.cost'), 0)) AS cost,
     substr('ses_ff50d39fbffe2mOJcly5gtxQyW', -8) AS session
   FROM ranked_calls
   WHERE call_number >= (
     SELECT MAX(call_number) - 9 FROM ranked_calls
   )
   ORDER BY call_number DESC;
Screenshot and/or share link

Zen pricing:
Image
Local sqlite pricing:

Image
Operating System

No response

Terminal

No response

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source file or test is named. Start by reproducing the issue with grok 4.6 or a GPT 5.6 model, then use the supplied SQLite query to compare stored call costs with Zen's pricing. Done means sessions crossing a context-pricing tier produce local estimates that match the applicable tiered costs.

Written by the indexing model from the issue text.

Assessment

Tech stack
sqlite, typescript
Domain
backend, databases
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
52/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.