Experiment with DeepSeek V4 Pro 0813 for Auto and Standard

[ross0x01]

Summary

• add HAC-68 server-side PostHog assignment for requests that already resolve to DeepSeek V4 Pro through HackerAI Auto or explicit Standard
• route treatment to deepseek/deepseek-v4-pro-0813 in direct requests and the durable Trigger Agent worker
• emit a privacy-safe custom exposure immediately before the first provider request and attach assignment to completion, reliability, latency, and cost analytics
• fail closed to control and fall treatment back through current DeepSeek V4 Pro before the existing Grok 4.6 / Kimi K3 recovery chain
• register the treatment model's OpenRouter pricing, cutoff metadata, text-only behavior, and focused coverage

Launch gate

Verified 2026-08-14 immediately before proceeding:

• HAC-64 is Done
• exact treatment slug deepseek/deepseek-v4-pro-0813 is healthy on DeepSeek, GMICloud, SiliconFlow, Novita, and Fireworks
• Cloudflare is listed but unhealthy and is not counted

Experiment

• Linear: HAC-68
• PostHog: experiment 418715
• Flag: deepseek_v4_pro_0813_model_v1
• Variants: control = deepseek/deepseek-v4-pro; test = deepseek/deepseek-v4-pro-0813
• Eligible surface: text-only Auto and Standard requests in Ask and Agent; existing image/PDF promotion remains unchanged
• The experiment remains draft and the flag inactive through merge/deployment. Launch follows internal production exposure verification.

Validation

• corepack pnpm typecheck
• corepack pnpm test — 358 suites / 3,664 tests passed
• staged ESLint and Prettier checks passed
• git diff --check

Manual verification after deployment

1. Keep experiment 418715 draft and confirm Auto/Standard text requests still use the current control route while the flag is inactive.
2. Temporarily allowlist an internal account, evaluate it to test, and send one text request in Auto and one in explicit Standard.
3. Confirm hac68_deepseek_v4_pro_0813_experiment_exposed fires immediately before the provider request with experiment_key, experiment_variant, configured_model, Auto/Standard selection, mode, and no user content.
4. Confirm provider attribution is deepseek/deepseek-v4-pro-0813, then remove the temporary allowlist before the planned broad 50/50 launch.

Summary by CodeRabbit

• New Features
  • Added support for the DeepSeek V4 Pro 0813 model across chat and agent experiences.
  • Added clear model labeling and support for recognized provider identifiers.
• Reliability Improvements
  • Added fallback routing to help maintain responses when the primary route is unavailable.
  • Improved handling of model selection and routing across supported experiences.
• Usage & Pricing
  • Added model-specific pricing and usage tracking.
• Quality Improvements
  • Added safeguards for unsupported multimodal tool-result scenarios.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
hackerai	Ready	Preview	Aug 14, 2026 3:15pm

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 6a3e44f0-5b5d-4402-8f22-abb995cccd79

📥 Commits

Reviewing files that changed from the base of the PR and between 844c19e and 1e64fbb.

📒 Files selected for processing (8)

• lib/ai/__tests__/providers.test.ts
• lib/ai/providers.ts
• lib/api/__tests__/chat-stream-helpers-fallback.test.ts
• lib/api/chat-handler.ts
• lib/api/chat-stream-helpers.ts
• lib/rate-limit/__tests__/token-bucket.test.ts
• lib/rate-limit/token-bucket.ts
• trigger/agent-long.ts

🚧 Files skipped from review as they are similar to previous changes (8)

• lib/api/tests/chat-stream-helpers-fallback.test.ts
• trigger/agent-long.ts
• lib/rate-limit/tests/token-bucket.test.ts
• lib/api/chat-handler.ts
• lib/rate-limit/token-bucket.ts
• lib/ai/tests/providers.test.ts
• lib/api/chat-stream-helpers.ts
• lib/ai/providers.ts

---

📝 Walkthrough

Walkthrough

Adds the DeepSeek V4 Pro 0813 model registry, experiment-based model selection, fallback and retry routing, pricing aliases, analytics context, exposure capture, and test coverage across chat and Agent Long flows.

Changes

DeepSeek V4 Pro 0813 rollout

Layer / File(s)	Summary
Model registry and pricing lib/ai/providers.ts, lib/ai/__tests__/providers.test.ts, lib/rate-limit/token-bucket.ts, lib/rate-limit/__tests__/token-bucket.test.ts	Registers the dated provider alias, metadata, DeepSeek classification, multimodal restriction, and existing DeepSeek V4 Pro pricing.
Experiment evaluation and analytics lib/experiments/deepseek-v4-pro-0813.ts, lib/experiments/__tests__/deepseek-v4-pro-0813.test.ts	Adds eligibility checks, PostHog flag evaluation, typed assignments, exposure capture, analytics context, and active-assignment matching.
Fallback and retry routing lib/api/chat-stream-helpers.ts, lib/api/__tests__/chat-stream-helpers-fallback.test.ts	Routes the dated model through DeepSeek V4 Pro, then the existing Grok and Kimi fallback chain.
Chat and Agent Long integration lib/api/chat-handler.ts, trigger/agent-long.ts	Applies experiment model assignments, propagates routing context, and records exposure during stream and task setup.

Estimated code review effort: 4 (Complex) | ~45 minutes

Merge Risk: ⚪ Minimal · up to 1e64f

The PR adds an inactive, controlled model experiment with fallback behavior and documented validation; no actionable merge-blocking risk remains beyond normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant ChatHandler
  participant DeepSeekExperiment
  participant PostHogClient
  participant ChatStreamHelpers
  ChatHandler->>DeepSeekExperiment: evaluate selected DeepSeek Pro model
  DeepSeekExperiment->>PostHogClient: return control or test variant
  DeepSeekExperiment-->>ChatHandler: return assigned model and context
  ChatHandler->>DeepSeekExperiment: capture experiment exposure
  ChatHandler->>ChatStreamHelpers: resolve fallback route

Loading

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely identifies the DeepSeek V4 Pro 0813 experiment for Auto and Standard requests.

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/deepseek-v4-pro-0813-experiment

---

_{Comment @coderabbitai help to get the list of available commands.}