[Bug]: Cost estimation does not recognize deepseek-flash (deepseek/deepseek-flash shows $0)

Deployment type

Official App

Area

Agents and runtimes

What happened?

Agents configured with model id deepseek/deepseek-flash (also reported as bare deepseek-flash) do not get cost estimates in Multica usage / runtime dashboards — cost shows as $0 / unrecognized, even though token usage itself is recorded.

Expected: deepseek-flash should resolve to the same (or explicitly maintained) pricing row as DeepSeek's current Flash SKU, so estimated cost is non-zero like deepseek-v4-flash.

Actual: Built-in pricing tables only know:

• Frontend MODEL_PRICING (packages/views/runtimes/utils.ts): deepseek-v4-flash, deepseek-v4-pro, deepseek-chat, deepseek-reasoner
• Server aliases (server/internal/metrics/pricing.go): regex deepseek-v4-flash|^deepseek-chat$|^deepseek-reasoner$ → deepseek:v4-flash

There is no entry / alias for deepseek-flash.

Quick check:

Model id	Priced today?
deepseek-v4-flash / deepseek/deepseek-v4-flash	yes
deepseek-chat / deepseek-reasoner	yes (aliased to v4-flash)
deepseek-flash / deepseek/deepseek-flash	no

models.dev currently lists both deepseek-v4-flash (DeepSeek V4 Flash) and deepseek-flash (DeepSeek V4.1 Flash) as separate ids with published rates — Multica's catalog has not caught up to the newer id.

Related prior work:

• #3204 / commit adding DeepSeek v4-flash pricing (aliases chat/reasoner only)
• #7606 (broader model-alias / automatic pricing feature)

Workaround today: Runtime UI Custom pricing (multica_runtime_custom_pricing local store) can fill rates manually — browser-local, not a first-party catalog fix.

Steps to reproduce

1. Create or update an agent with model deepseek/deepseek-flash (or deepseek-flash).
2. Run tasks so token usage is recorded.
3. Open Runtime / Usage (or issue usage) cost views.
4. Observe estimated cost is $0 / model treated as unpriced, while the same flow with deepseek-v4-flash estimates correctly.

Suggested fix

1. Add deepseek-flash to frontend MODEL_PRICING and server pricing aliases (map to the official Flash / V4.1 Flash rates, or alias onto the existing deepseek-v4-flash row if product intends them equivalent).
2. Ideally keep aliases in sync with models.dev / DeepSeek docs so renames like v4-flash → flash do not silently drop cost again (#7606).

Additional context

• Observed on Official App with agents whose model field is literally deepseek/deepseek-flash.
• Temporary mitigation: set custom pricing for deepseek-flash in the Runtime usage custom-pricing dialog.