[Feature]: Show average tokens/sec in Usage by model and provider #6309

Area

Dashboard → Usage

What are you trying to accomplish?

Compare average output tokens/sec across models and providers for the selected usage period, without inspecting individual requests.

What prevents this today?

Logs already shows per-request end-to-end throughput and estimated decode speed (#4038, implemented in #4166). Usage shows request counts, tokens, cache metrics and estimated costs, but its summary and Models/Providers tables have no average tokens/sec. Confirmed in v2.73.0 and current main/dev source.

What should OpenCodex do?

• Add average output tokens/sec to the Usage summary and Models/Providers tables, respecting the selected date range and filters.
• Clearly distinguish end-to-end throughput from estimated decode speed after the first output; document the aggregation method.
• Aggregate from usage history on the server, rather than the limited recent Logs buffer.
• Exclude requests without valid token/timing data, show the contributing request count, and display unavailable when no samples qualify.

Example usage or interface

Select 7d → Models and compare each model's average tokens/sec alongside its request count and token usage; switch to Providers for the provider-level comparison.

Additional context

#4038 explicitly scoped its implementation to Logs. This requests aggregate throughput in Usage.

Checks

• I searched existing issues and documentation.
• This request describes a concrete OpenCodex workflow rather than merely naming a desired technology.
• I removed secrets and personal data.