Cost-Per-Pass, Not Cost-Per-Call#

Practitioners price LLMs by the per-token rate on the provider’s pricing page. For autonomous agents, that number is misleading. Two layers of indirection sit between the per-token rate and the cost you actually pay to get work done: variable prompt sizes turn per-token into per-call, and variable pass rates turn per-call into per-pass. Each layer can invert the ranking.

For autonomous fleets where failed attempts trigger reviewer cycles, retries, and reputational drag, cost-per-pass is the only metric that ranks models correctly. This article shows how to compute it, when it dominates, and where the cheapest-per-token model becomes the most expensive in production.

DeepSeek V4 Operational Quirks#

DeepSeek V4 ships two models behind one OpenAI-compatible API: V4-Pro (reasoning) at $1.74/M input / $3.48/M output and V4-Flash (chat) at $0.28/M input / $1.10/M output. Until 2026-05-31 V4-Pro carries a 75% discount, putting it at $0.435/M input — cheap enough to use as a heavy-tier coding model. After that, the cost steps up 4×.

The two models live on the same endpoint but want very different things. V4-Pro behaves like a reasoning model (thin prompts, reasoning_content echo required, tool_choice restrictions). V4-Flash behaves like a chat model (rich prompts win dramatically; rejects nothing). Confuse them and your matrix lights up red.

LLM Adapter Audit Checklist#

When you wrap an OpenAI-compatible LLM provider (Moonshot, DeepSeek, xAI, Together, Fireworks, OpenRouter, vLLM, anything else that exposes POST /v1/chat/completions) in a Go HTTP client, the same ten bug classes show up. They all silently degrade or break the agent — none of them crash loudly. Each was observed in production across at least one of xAI, DeepSeek, or Moonshot during a two-week audit period.

This checklist is the audit. Run it against any new adapter before shipping. Each entry is Symptom → Cause → Fix with a code shape you can grep your repo for.

OFAT Matrix LLM Tuning#

When a new provider or model lands and you have to decide what temperature, max_tokens, tool_choice, prompt-shape, and turn budget to ship in production, the default is to pick by hunch. Read the model card, copy a partner adapter’s defaults, ship. A week later you find out reasoning_effort=high doubled cost for no quality gain, max_tokens=2048 silently truncated half your tier-3 runs, and the “prompt-rich” pattern you copied from grok-4.3 actively hurts kimi.

Reasoning-Model Tuning Asymmetry#

Practitioners assume “better prompt = better output”. For one model class, that assumption is correct. For the other, the same prompt makes things measurably worse. This article documents the asymmetry, names the dividing line, and gives you a 4-cell test to confirm it on your own canary before you commit to a prompt.

The asymmetry is empirical, not theoretical. It shows up cleanly across four independent OFAT (one-factor-at-a-time) matrices run between 2026-05-18 and 2026-05-20: sonnet POC, grok matrix v1+v2, deepseek matrix v1, kimi matrix v1.

The d4-rich Prompt Pattern#

Non-reasoning chat models (deepseek-V4-Flash, grok-4.3, kimi with thinking disabled) collapse on multi-file refactor tasks when given thin or baseline prompts. Pass rates of 0-33% on canaries that reasoning models clear at 67-100%. The cheap fix is a three-part prompt addendum: completion checklist, callsites-exhaustively-updated rule, and verify-before-push instruction. Drop it into the system prompt of a non-reasoning model and the canaries go green. Drop it into a reasoning model and you pay 12× more for 0% quality improvement.