Run an Autonomous Coding Agent
Best when the model will drive an agent loop with tools (shell, editor, tests) for extended periods. Heavily weights execution, reliability, tool calling, and long autonomous runs. Recommendations must never hide the agent scaffold; evidence without a named harness is weak for this quest.
Fit score ≠ absolute IQ
- Fit score is a weighted blend of this Quest’s attribute/skill weights (and optional cost/tempo soft prefs). Higher = better match for this job among scored Builds.
- Attribute inputs are peer-public scores (roughly 40–95 within the current peer set). A low STR here can still be a strong absolute model — it just ranks lower among current scored peers on that axis.
- Decisive means the leader is ≥4 fit points ahead of #2. Smaller gaps → pick on price, Party stack, or taste.
Peer set today: 6 scored Builds. Adding another re-ranks peers — re-run the engine after new evidence.
Best fit
Decisive lead on weighted fit. Strongest blend on STR + CON. OpenAI flagship Build for autonomous coding and professional workflows when ecosystem tools and max reasoning justify $5/$30.
Struggle · Latency-first everyday chat (use lower effort or Luna/Terra)
Suggested multi-model stack
Most jobs want a Party — not one model forever. Primary handles the Quest; ceiling covers hard escalations; value handles volume and drafts.
Decisive Quest Fit lead — default seat for this job.
Peer INT 85 — escalate hard reasoning or research here.
Cheapest list stack among remaining candidates ($0.435/$0.87) for volume or drafts.
What we weight
Relative importance of each input in the Fit formula.
- Always disclose agent harness on evidence and in recommendation copy.
- Never treat bare-chat scores as autonomous-agent proof.
Contribution breakdown
Peer-public input × Quest weight. Bars show weight mass; numbers show the score fed into Fit.
All scored candidates
Decisive lead on weighted fit. Strongest blend on STR + CON. OpenAI flagship Build for autonomous coding and professional workflows when ecosystem tools and max reasoning justify $5/$30.
Struggle · Latency-first everyday chat (use lower effort or Luna/Terra)
10.2 fit points behind GPT-5.6 Sol · API · Max effort. Still solid where STR / CON matter. The job is hard enough that a wrong answer costs more than Opus tokens.
Struggle · Everyday chat and low-stakes drafts (use Sonnet/Flash)
17.8 fit points behind GPT-5.6 Sol · API · Max effort. Still solid where STR / CON matter. You want near-frontier coding and agentic work with strong price and speed.
Struggle · EU-only deployments until regional GA
29.5 fit points behind GPT-5.6 Sol · API · Max effort. Still solid where STR / CON matter. The quest is research, multimodal, or long-context — not pure SWE ceiling at any price.
Struggle · CON · Reliability is a relative soft spot (42 peer public)
37.3 fit points behind GPT-5.6 Sol · API · Max effort. Still solid where STR / CON matter. You want open-frontier capability at roughly an order of magnitude below Sol/Opus list prices.
Struggle · CON · Reliability is a relative soft spot (40 peer public)
38.0 fit points behind GPT-5.6 Sol · API · Max effort. Still solid where STR / CON matter. You want one primary Build for coding and writing that stays pleasant under human supervision, holds constraints, and rarely needs the top-tier price.
Struggle · STR · Execution is a relative soft spot (40 peer public)
Capability snapshot
Peer public ranks feed Fit; raw aggregates stay absolute evidence fusion. Full sheet: Codex →
What matters most
- Multi-step tool sequencing without getting stuck
- Recovery after failed commands
- Honest completion claims backed by verification
Common failure modes
- Infinite loops and verbose thrashing
- Claiming success without green tests
- Harness-inflated scores misread as raw model ability