Before They Can Solve: Predicting Post-Training Coding-Agent Performance from Base Models
Three screens rank base models by likely coding-agent performance after expensive post-training.
The paper asks how to predict which base checkpoint merits expensive agentic post-training for coding. End-to-end pass@K is a poor fit because many base models cannot emit the tool calls needed to finish a task. Replaying successful post-trained trajectories, the authors locate the decisive step that first makes tests pass, then score it with Decisive-Action bits-per-byte, a patch multiple-choice test, and prefix-conditioned pass@K. Across ten public base and post-trained pairs, all three screens rank models in close agreement with post-trained SWE-bench Verified pass@1.
- Decisive step is the first patch that flips tests from fail to pass
- Screens are Decisive-Action BPB, Patch MCQ, and prefix-conditioned pass@K
- Base models need not drive the harness from a cold start
- Ten base and post-trained pairs track SWE-bench Verified pass@1
Full article276 words · extracted from arxiv.org · click to collapse
How can we predict which base checkpoint is worth an expensive round of agentic post-training? End-to-end pass@$K$ tests whether successful behavior already appears in a base model's distribution, but it is a poor fit for agentic coding: many base checkpoints cannot reliably produce the well-formed tool invocation required to complete a task end-to-end. Single-shot or short-horizon tasks avoid these tool-calling failures by collapsing a multi-step interaction into a fixed prompt and a single patch, but they sidestep the core capability we care about: maintaining coherent state over many tool-using steps as the repository evolves. To bridge this gap, we treat successful post-trained agent trajectories as a lookahead signal of base-model potential. Replaying each trajectory and rerunning tests after every code-changing step identifies the decisive step: the first step whose cumulative patch flips the repository from failing to passing, certifying that the recorded action solves the task given the prior context. Motivated by a coverage principle for agentic traces, we build three screens at this step that do not require a base checkpoint to drive the harness from a cold start: (i) Decisive-Action BPB (bits per byte) measures the probability mass on the certified action, (ii) Patch MCQ tests the checkpoint's choice between that action and alternatives rejected by the same verifier, and (iii) prefix-conditioned pass@$K$ evaluates support for functionally-correct generations and credits any continuation that the tests accept. Across ten pairs of public base and post-trained models, all three screens rank the cohort in close agreement with post-trained SWE-bench Verified pass@$1$. As our methods need only a benchmark's successful trajectories and its verifier, they can be applied to turn future agentic coding benchmarks into base-model evaluations.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.10478