Electricity Bench

antigravity-cli + gemini-3.7-flash

antigravity-cli harness · Google AI Pro

COVERALL

Clears tier 4 of 5 (partial at tier 5) on Real-world issues

Capability mean across 2 graded suite families.

capability
73%
plan cost per run
9.0% of Google AI Pro
median task time
2m 25s
reliability
100%
12345
What the tiers mean

"Clears" marks the highest rung where every task at it and below was solved. Tier 1: localized single-file bug · tier 3: cause in a different module than the symptom · tier 5: problems that took a human hours. A cleared rung solved every task on it; a partial rung solved some. The headline stops at the last unbroken rung, so a solved tier above a half-solved one counts only as partial.

Clears level 2 of 5 (partial at levels 3, 4 and 5) on Vibe coding v1 - the vaguest report level at which it still fixed every bug, having also cleared every richer level below
12345
What the levels mean

Level 1: a full brief naming the surface and the acceptance · level 3: the report as filed · level 5: a vague vibe report in a non-technical user's words. Context falls as the level rises, so vaguer is harder. A cleared level fixed every bug at it and below; a partial level fixed some. The headline stops at the last unbroken level, so a fix above a half-cleared level counts only as partial.

11 of this suite's tasks were credited rather than run: on a suite that asks the same problem at several levels of detail, solving it from the vaguest description credits the more detailed ones instead of asking again.

Per-suite results

C
Real-world issues v3
73%
plan cost per run
9.0% of Google AI Pro
median task time
2m 25s
reliability
100%
tier 1tier 2tier 3tier 4tier 5

13 of 15 tasks solved - grouped by tier, easiest first; hover a square for its time and turns.

Run on antigravity-cli 1.1.12, model gemini-3.7-flash (--effort medium), week 2026-W33.

One run of this suite consumes 54.0% of Google AI Pro's 5h window and 9.0% of the weekly limit.

How the plan cost was measured

Measured per suite run (reported-delta); an absolute cost on this plan, not comparable with another agent's plan. Measured on a benchmark-only account (Dedicated benchmark Google account ebbench.runner@gmail.com on eb-runner; no interactive or IDE use shares the AI Pro weekly Gemini window.).

C
Vibe coding v1
72%
plan cost per solve
0.7% of Google AI Pro
median task time
3m 37s
reliability
100%
level 5level 4level 3level 2level 1issue aissue bissue cissue dissue e

21 of 25 tasks solved (11 credited without running) - each row is one bug, vaguest report first and context growing to the right; hover a square for its time and turns.

Run on antigravity-cli 1.1.13, model gemini-3.7-flash (--effort medium), week 2026-W33.

One good solve on this suite consumes, on average, 4.0% of Google AI Pro's 5h window and 0.7% of the weekly limit.

How the plan cost was measured

Counted over each issue's first solve (5 solves); the walk's failed harder cells and confirmation runs are the benchmark's own search cost and are excluded (full walk: 8.7% over 14 cells). Measured per cell run (reported-delta); an absolute cost on this plan, not comparable with another agent's plan. Measured on a benchmark-only account (Dedicated benchmark Google account ebbench.runner@gmail.com on eb-runner; no interactive or IDE use shares the AI Pro weekly Gemini window.).

Task outputs from these runs are withheld: publishing them would reveal the private suite. Per-task scores, times and turns are in the chips above.

Trajectory

One point per grading week - a re-run within a week shows only its latest execution. Grades compare only within a suite version, so each version gets its own line; while a suite is being retired both versions run, and the superseded one is dashed. The filled point is the run the scorecard shows today. Click a point or week for plan usage. All agents over time →

Suite
Real-world issues · 2 weeks graded · latest 2026-W33
50%60%70%80%2026-W322026-W33New registration - model gemini-3.6-flash -> gemini-3.7-flash (released 2026-08-13), same harness, plan and effortantigravity-cli + gemini-3.7-flash · 2026-W32 · Aug 5, 2026 · 60% · Cantigravity-cli + gemini-3.7-flash · 2026-W33 · Aug 13, 2026 · 73% · C

Dotted = a new model version took the line over (grok-4.5 → grok-4.6): the points either side were graded by different models. Plan or settings re-registrations join solid.

Weekly heads

Newest week first. Click a row for usage.

WeekGradeCapabilityClearsTask timeUsage
2026-W33 · v3nowC73%tier 4 · p52m 25s9.0% of Google AI Pro
2026-W32 · v3C60%tier 4 · p52m 32s19.6% of Google AI Pro

Frequently asked questions

Is gemini-3.7-flash worth it on Google AI Pro?

On our private real-world suite (real-world@v3), gemini-3.7-flash earned grade C with 73% capability, clearing every problem up to tier 4 of 5 - a cross-cutting problem needing broad codebase orientation and landing some tier 5 problems. It runs on Google AI Pro. One full suite run consumed 9.0% of the plan's week window. Measured 2026-08-13.

How good is gemini-3.7-flash at real-world coding?

gemini-3.7-flash clears tier 4 of our five-tier real-world ladder - every task at that tier and below - where tier 1 is a localized single-file bug and tier 5 is a problem that took a human engineer hours. It solves some but not all tasks up at tier 5. Its capability score on real-world@v3 is 73% (grade C). Last benchmarked 2026-08-13.

Compare

In the tools directory