Pimp My IDE / Garage Dispatch
← Back to the garage
September 10, 2026 · coding models / effort / operating policy

Your coding agent needs an effort gearbox.

SWE-2 arrives with medium, high, and max as points on a cost–performance curve, not three moods for the same robot. That is the useful release note: reasoning depth is becoming an operating decision. Stop pinning every task in top gear.

The take: choose effort from task risk, ambiguity, and proof burden—not benchmark swagger. A typo does not deserve a moon launch. A destructive migration does not deserve economy mode. The control belongs in the workflow, with an escalation rule and a receipt.
Shift the effort gearbox ↓

The headline is not 92.8.

Cognition reports that SWE-2 scores 92.8% on Terminal-Bench 2.1 and 27.3% on Terminal-Bench 4. It also reports 50.0% on FrontierCode 1.1 Main, one point below Fable 5.1 at 64% lower cost. Those are vendor-published measurements, not independent guarantees for your repository.[1]

The more actionable number is behavioral: Cognition says SWE-2 medium used 53 mean steps on its FrontierCode runs, versus 80 at high and 98 at max. Medium reportedly reached its first real edit in a median 18 steps, compared with 48 for SWE-1.7.[1] The release is selling a model, but the engineering idea survives the sales pitch: deeper is a budget with a shape.

“Max” is not a quality setting. It is a purchase order for more search.

The effort knob is spreading.

OpenAI’s current reasoning documentation exposes model-dependent effort values from none through max. It describes lower effort as favoring speed and token use, while higher effort spends more work on complex reasoning; it explicitly recommends evaluating whether the extra latency and cost earn a benefit.[2]

Cognition says SWE-2 trains multiple effort levels in one reinforcement-learning run with a cost penalty tuned to each point on the base model’s Pareto curve. It says SWE-2 is post-trained from Kimi K3.[1] Moonshot’s Kimi K3 repository identifies K3 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion active parameters and publishes weights under its own Kimi K3 License.[3]

Different vendors expose different gears and train them differently. Do not pretend the labels are portable units. “High” is not horsepower. It is a product-specific request to spend more inference on the problem.

Benchmarks are the dyno, not the street.

The Hacker News discussion immediately centered on the gap between SWE-2’s Terminal-Bench 2.1 and Terminal-Bench 4 scores. Commenters dispute whether that gap indicates benchmark saturation, contamination, different difficulty, or weak generalization.[4] The argument is useful precisely because the post’s clean table cannot settle it.

Run your own mixed garage route: small edits, ambiguous bugs, ugly migrations, tool failures, stale tests, and tasks where the correct move is to stop. Record success, review burden, first-edit latency, total cost, and rollback pain. A model that wins a public terminal benchmark can still make your monorepo miserable.

Wire the shift points into policy.

  1. Start from consequence. Low-risk, reversible work can begin in a cheaper gear. Security, migrations, money, and destructive operations start higher.
  2. Add ambiguity. Unknown ownership, unclear acceptance criteria, or weak tests justify more exploration—or a human question, not infinite thought.
  3. Name the shift-up trigger. Escalate on failed verification, contradictory evidence, repeated edits, or contact with a protected boundary.
  4. Name the shift-down trigger. Once the plan is proven and the remaining work is mechanical, stop paying for philosophical suspension.
  5. Keep proof independent. More reasoning does not waive tests, review, provenance, or rollback. It only changes how much work the model spends before handing you the claim.
Interactive makeover / reasoning control

Effort Gearbox.

Traditional purpose replaced: a hidden “thinking” dropdown. Better version: task consequence, ambiguity, and proof burden sit beside the selected gear; the panel recommends a shift and prints the exact operating receipt. Recommendation is a teaching proxy, not a model evaluation.

Select the working gear

Use the radio group by keyboard or pointer. Then set the road load. The box will call out an under-geared or over-geared run.

Reasoning effort
2
2
3

Shift telemetry

Medium matches this moderate task load. Keep the checks outside the model.

MEDIUMselected gear
MEDIUMrecommended start
HOLDshift call
Open the four-source dyno sheet
[1] Cognition — “Introducing SWE-2: Pushing the Pareto Frontier,” September 10, 2026: vendor benchmark, cost, behavior, training, and availability claims. [2] OpenAI API documentation — reasoning effort: model-dependent effort values, tradeoffs, and workload guidance. [3] Moonshot AI — Kimi K3 repository: architecture summary, parameter counts, context, weights, and license link. [4] Hacker News item 49645443 — live practitioner debate around benchmark saturation, difficulty, generalization, and cost; reaction, not authority.

Source boundary: the article treats the launch numbers as claims by their publisher and the discussion as discussion. The gearbox is an operating-design pattern derived from those sources; its score is not a benchmark, safety rating, or recommendation for any particular vendor.