Problem
The gpt-6 family codex models (Sol, Luna, Astra) hardcode a 272K context window via injectCodexGpt6Models() in generate-models.ts. This blocks these models from inheriting correct limits from models.dev.
Evidence
- injectCodexGpt6Models() at line ~87 hardcodes contextWindow: 272_000 and maxTokens: 128_000
- models.dev API already carries correct metadata (openai/gpt-6.1-sol: 1.05M context, 922K input, 128K output)
- The hardcoded values prevent models.dev from being the source of truth for context limits
Impact
Models always report 272K context regardless of actual API limits in models.dev, misleading users about available context.
Solution
Remove the 272K hardcoding and let models inherit from models.dev through the existing applyGlobalModelsDevFallback mechanism using UNK_CONTEXT_WINDOW and UNK_MAX_TOKENS sentinels.
—
[repo owner's gaebal-gajae (clawdbot) 🦞]
Problem
The gpt-6 family codex models (Sol, Luna, Astra) hardcode a 272K context window via
injectCodexGpt6Models()in generate-models.ts. This blocks these models from inheriting correct limits from models.dev.Evidence
Impact
Models always report 272K context regardless of actual API limits in models.dev, misleading users about available context.
Solution
Remove the 272K hardcoding and let models inherit from models.dev through the existing applyGlobalModelsDevFallback mechanism using UNK_CONTEXT_WINDOW and UNK_MAX_TOKENS sentinels.
—
[repo owner's gaebal-gajae (clawdbot) 🦞]