Very high input token usage with Cerebras (GLM 4.7)

Description

Hi,
Token usage as reported (and billed) by Cerebras is completely different from the usage report in opencode.

This is the usage report in opencode for the past day:

┌────────────────────────────────────────────────────────┐
│                      MODEL USAGE                       │
├────────────────────────────────────────────────────────┤
│ cerebras/zai-glm-4.7                                   │
│  Messages                                           62 │
│  Input Tokens                                   168.6K │
│  Output Tokens                                   14.3K │
│  Cost                                          $0.0000 │
└────────────────────────────────────────────────────────┘

For the same period, Cerebras reports 63 total requests, 2.56M total tokens (2.55M input tokens and 14.39k output tokens).

This is a 15x difference!

In normal usage, with a single opencode instance I hit the 500k tokens/min limit a few times in about 10 minutes of usage.

Plugins

opencode-antigravity-auth@latest, @mohak34/opencode-notifier@latest

OpenCode version

1.1.36

Steps to reproduce

Use opencode with cerebras/zai-glm-4.7 model
Notice very high input token usage

Screenshot and/or share link

No response

Operating System

Ubunut 25.10

Terminal

Alacritty and vscode terminal

This issue might be a duplicate of existing issues. Please check:

Token count is way off #5951: Token count is way off
Mismatch between token spent and billing against my account #8945: Mismatch between token spent and billing against my account
Excessive input token usage with Sonnet on AWS Bedrock #10066: Excessive input token usage with Sonnet on AWS Bedrock
High Token Consumption Observed with Opus 4.5 in OpenCode #6728: High Token Consumption Observed with Opus 4.5 in OpenCode
When using the cerebras api i am frequently getting: "Tokens per minute limit exceeded - too many tokens processed" #7463: When using the cerebras api i am frequently getting 'Tokens per minute limit exceeded'
GLM 4.7 from Cerebras not working #8005: GLM 4.7 from Cerebras not working
web and tui reports token usage differently. #3322: web and tui reports token usage differently

Feel free to ignore if none of these address your specific case.

Any updates on this? I encountered similar issues.

Is it possible cerebras is doing something weird? I noticed that glm costs skyrocket like crazy with only a few minutes of usage, but using gpt-oss-120b on cerebras didn't skyrocket even though I used it way more, and, glm4.7 on openrouter didn't skyrocket like it did on cerebras for similar usage.

I don't have that problem with other AI coding agents using Cerebras.

I heard that cerebras counts tokens in a really weird way, and I bet we aren't treating it correctly

Cerebras counts cached tokens as normal tokens, so this may be it.

To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

Very high input token usage with Cerebras (GLM 4.7) #10851

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Very high input token usage with Cerebras (GLM 4.7) #10851

Description

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on Jan 27, 2026

Sandipbgn commented on Jan 29, 2026

icculp commented on Feb 3, 2026

azais-corentin commented on Feb 3, 2026

rekram1-node commented on Feb 3, 2026

azais-corentin commented on Feb 10, 2026

github-actions commented on Apr 12, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions