feat(app): add DeepSeek V4.1 Flash to the profit estimators with DeepSeek Flash list-price defaults / 利润估算器新增 DeepSeek V4.1 Flash 并默认采用 DeepSeek 官方定价

[functionstackx]

Summary

Adds DeepSeek V4.1 Flash (DeepSeek V4.1 Flash 552B, registered in #1109) to /profit-estimator and /profit-estimator-per-gigawatt, plus the /zh mirrors, following the pattern from #1007 (DeepSeek V4 Pro).

• Routes: /profit-estimator/deepseek-v41-flash and /profit-estimator-per-gigawatt/deepseek-v41-flash (PROFIT_ESTIMATOR_MODELS in model-routes.ts; the slug is the site-wide deepseek-v41-flash from the compare registry). Kimi K3 stays the bare-path default. Models outside the allow-list still 404 and stay out of the sitemap. Selector order is now DeepSeek V4 Pro, DeepSeek V4.1 Flash, Kimi K3, MiniMax M3, GLM 5.2.
• Per-model defaults (profitModelDefaults(Model.DeepSeek_V4_1_Flash)):
  • 125 tok/s/user, the speed DeepSeek serves Flash at.
  • DeepSeek list price $0.3 input / $0.006 cached / $1.2 output per M tok, the peak-hour rate for deepseek-flash on the DeepSeek pricing page (off-peak is 50% of that: $0.15 / $0.003 / $0.6; peak is 01:00-04:00 and 06:00-10:00 UTC on weekdays). The same page says deepseek-v4-pro requests route to V4.1 Flash at the Flash price from Sep 14 until V4.1 Pro ships. OpenRouter and Custom remain one click away; the caption links the DeepSeek pricing page when the list price is in force.
  • 0% model license fee. The V4.1 Flash weights ship under the MIT license, so there is no lab cut to model (V4 Pro stays on 5%).
• Data: the live /api/v1/availability has no dsv41flash agentic rows yet (the model entered the fleet AgentX-only via InferenceX#2961), so the in-page selector, which intersects the allow-list with models that have agentic data, will not list it in production until those rows land. The slugged routes exist now and will populate without a code change. Any curve whose P90 interactivity range does not straddle 125 will list under Not priced, per the existing clamp-skip behaviour.
• Model switch re-seeds interactivity, price source, and license fee in every direction (V4 Pro 24/list/5%, Flash 125/list/0%, Kimi 45/OpenRouter/30%).
• Docs: docs/tco-calculator.md profit-estimator section.

Tests

• Unit (vitest): profitModelDefaults(Model.DeepSeek_V4_1_Flash) and its list to revenue-pricing conversion. modelRoutesForTab expects [DeepSeek V4 Pro, DeepSeek V4.1 Flash, Kimi K3, MiniMax M3, GLM 5.2] in MODEL_ROUTES order, the deepseek-v41-flash slug, and Kimi K3 as the tab default. bun run typecheck, bun run lint, bun run fmt, bun run check:typography, and the full vitest suite pass locally (the one pre-existing benchmark-transform.test.ts failure reproduces on master and is unrelated, same as noted in feat(app): add DeepSeek V4 Pro to the profit estimators with DeepSeek list-price defaults / 利润估算器新增 DeepSeek V4 Pro 并默认采用 DeepSeek 官方定价 #1007).
• E2E (Cypress): fixtures now emit dsv41flash agentic rows keyed by ?model=DeepSeek-V4.1-Flash. The wide fixture curve tops out at 130 tok/s/user so B200/B300/GB300/MI355X price at 125; the H200 curve tops at 38, so the new spec asserts H200 lands under Not priced. New specs cover the /deepseek-v41-flash route defaults, caption, and pricing-source link; Custom seeding from the list price; V4 Pro to Flash to Kimi model-switch re-seeding with the expected pathnames; and the /zh mirror label (输入：$0.3, DeepSeek 官方定价). The OpenRouter stub gains a deepseek/deepseek-v4.1-flash row and the selector assertion now expects five models. Ran profit-estimator.cy.ts locally in Electron: 45/45 pass on the final commit (an earlier run hit an order-dependent flake in per GW › switches to custom $/GPU/hr when a caption badge is edited that also reproduces on master and passes in isolation).

Screenshots from the fixture-backed local run are posted in the Slack thread.

中文说明

在 /profit-estimator 与 /profit-estimator-per-gigawatt（含 /zh 镜像）新增 DeepSeek V4.1 Flash（DeepSeek V4.1 Flash 552B，模型本身由 #1109 注册），路径为 /deepseek-v41-flash，沿用 #1007（DeepSeek V4 Pro）的模式。Kimi K3 仍为默认模型，其他模型继续返回 404。

• 按模型的默认值：默认 125 tok/s/user（DeepSeek 官方 API 对 Flash 的实际服务速度），并采用 DeepSeek 官方定价：输入 $0.3 / 缓存输入 $0.006 / 输出 $1.2（每百万 token），即官方定价页上 deepseek-flash 的高峰时段价格（非高峰为 5 折：$0.15 / $0.003 / $0.6）。OpenRouter 与自定义仍可一键切换。模型许可费默认 0%（权重采用 MIT 许可，无需计入实验室分成；V4 Pro 仍为 5%）。
• 数据：线上 /api/v1/availability 目前还没有 dsv41flash 的 agentic 数据（模型通过 InferenceX#2961 仅以 AgentX 形式进入 fleet），因此页面内的模型选择器在数据落地前不会显示该模型；带 slug 的路由已就位，数据到达后无需改代码。
• 切换模型时会重置交互性目标、价格来源与许可费。
• 同步更新 docs/tco-calculator.md。

测试：新增 vitest 单元测试（默认值、官方定价换算、路由白名单与 slug）；Cypress fixture 新增 dsv41flash agentic 数据并按 ?model= 参数返回，新增路由默认值（H200 在 125 tok/s/user 下列为未计价）、自定义价格初始化、V4 Pro → Flash → Kimi 切换重置与中文镜像的 e2e 用例。本地已通过 typecheck / lint / fmt / typography、完整单元测试，以及 profit-estimator.cy.ts（45/45）。

---

Note

Low Risk
Additive profit-estimator allowlist and defaults with tests and docs; no auth or shared calculator math changes.

Overview
Adds DeepSeek V4.1 Flash to the profit estimator tabs (/profit-estimator and /profit-estimator-per-gigawatt, including /zh), using the existing per-model route pattern.

The model is included in PROFIT_ESTIMATOR_MODELS with slug /deepseek-v41-flash and its own dsv41flash data bucket (separate from V4 Pro). profitModelDefaults opens Flash at 125 tok/s/user, DeepSeek peak list pricing ($0.30 / $0.006 cached / $1.20 per M tok), and 0% model license fee (MIT weights). Switching models still re-seeds interactivity, price source, and license fee; SKUs whose agentic curves do not reach 125 tok/s/user stay not priced (no extrapolation).

Coverage includes unit tests for defaults and routing, Cypress fixtures and specs (selector now five models, H200 excluded at 125, V4 Pro → Flash → Kimi), sitemap test comment, and docs/tco-calculator.md.

^{Reviewed by Cursor Bugbot for commit 6a40c96. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
inferencemax-app	Ready	Preview	Sep 11, 2026 3:11am UTC