---
type: "post"
slug: "ox-alpha-tops-opencode-usage-data"
title: "Ox Alpha tops OpenCode usage data"
description: "An unattributed model called ox-alpha took the top of OpenCode usage with 36T tokens at zero recorded spend, showing that free access plus a 94% cache ratio moves usage before provenance does."
retrieval_nugget: "An unattributed model called ox-alpha took the top of OpenCode usage with 36T tokens at zero recorded spend, showing that free access plus a 94% cache ratio moves usage before provenance does."
published_at: "2026-08-30"
updated_at: "2026-09-12"
record_date: "2026-08-25"
date_kind: "discovered_at"
topics: ["agents","ai","ai-software","coding-agents"]
entities: ["opencode.ai"]
source_url: "https://opencode.ai/data/unknown/ox-alpha"
source_title: "ox-alpha Usage, Cost & Rank"
source_domain: "opencode.ai"
source_terms: ["Alpha","tops","OpenCode","usage","ox-alpha"]
summary_word_count: 185
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:ox-alpha-tops-opencode-usage-data"
status: "published"
source_urls: ["https://opencode.ai/data/unknown/ox-alpha"]
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:176f82bae8a9b1596ed52e70bfa833da264f9cde8f0506cd44dd415080701db1","reviewed_at":"2026-09-12T10:00:00Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":2}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"OpenCode's public usage data puts a model called ox-alpha at rank one for the week with 8.9% of roughly 2M observed volume, 36 trillion tokens between 1 July and 25 August, 425,000 unique users and just over 11 million completed sessions.","observed_facts":[{"text":"ox-alpha is ranked first with 8.9% token share, 36 trillion tokens, 425,000 unique users and over 11 million completed sessions.","source_urls":["https://opencode.ai/data/unknown/ox-alpha"]},{"text":"Recorded total spend and average cost per session are both $0.00 with a 94% input cache ratio.","source_urls":["https://opencode.ai/data/unknown/ox-alpha"]}],"mechanism":"Total spend and average cost per session are both $0.00, and the average session consumes 3.2 million tokens at a 94% input cache ratio.","why_now":"The vendor fields — context, output, knowledge, release and inputs — are all recorded as Unknown.","implications":["A free, heavily cached model with very long sessions accumulates token share faster than a paid model can, so the leaderboard measures distribution economics rather than capability."],"evidence_boundary":"The data is OpenCode's own observed traffic, the model's provenance is unstated, and the geography breakdown records where tokens were spent rather than who spent them.","watch_conditions":["The condition to watch is whether ox-alpha keeps its share once a price is attached, because a zero-cost model leaving the top of the table would confirm the ranking tracked spend rather than preference."],"related_records":[{"url":"https://newruntime.com/posts/model-routing-and-pricing-update-glm-deepseek-grok-qwen","relation":"Prior coverage of pricing and routing shifting model usage."},{"url":"https://newruntime.com/topics/coding-agents","relation":"Topic hub where coding-agent model choice is tracked."}]}
routes: {"html":"https://newruntime.com/posts/ox-alpha-tops-opencode-usage-data/","markdown":"https://newruntime.com/posts/ox-alpha-tops-opencode-usage-data.md","json":"https://newruntime.com/posts/ox-alpha-tops-opencode-usage-data.json"}
---

# Ox Alpha tops OpenCode usage data

## Retrieval answer

An unattributed model called ox-alpha took the top of OpenCode usage with 36T tokens at zero recorded spend, showing that free access plus a 94% cache ratio moves usage before provenance does.

OpenCode's public usage data puts a model called ox-alpha at rank one for the week with 8.9% of roughly 2M observed volume, 36 trillion tokens between 1 July and 25 August, 425,000 unique users and just over 11 million completed sessions. Its nearest peer, deepseek-v4-flash, sits at 28 trillion tokens. The vendor fields — context, output, knowledge, release and inputs — are all recorded as Unknown.

Two numbers explain the ranking better than the rank does. Total spend and average cost per session are both $0.00, and the average session consumes 3.2 million tokens at a 94% input cache ratio. A free, heavily cached model with very long sessions accumulates token share faster than a paid model can, so the leaderboard measures distribution economics rather than capability.

That is the same distortion visible whenever pricing moves first, as in the [model routing and pricing update across GLM, DeepSeek, Grok and Qwen](https://newruntime.com/posts/model-routing-and-pricing-update-glm-deepseek-grok-qwen/), and it is a reason usage dashboards belong next to, not instead of, evaluation evidence in the [coding agents hub](https://newruntime.com/topics/coding-agents/).

The data is OpenCode's own observed traffic, the model's provenance is unstated, and the geography breakdown records where tokens were spent rather than who spent them. The condition to watch is whether ox-alpha keeps its share once a price is attached, because a zero-cost model leaving the top of the table would confirm the ranking tracked spend rather than preference.