AgentTested hands-on review
tool: Poolside Laguna S 2.1
slug: laguna-s-2-1
date: 2026-10-21
score: 6/10

Coverage: Same 4-task battery. Only 4 of 8 task runs completed because of rate limits.
What we tested: Same 4-task battery. Only 4 of 8 task runs completed because of rate limits.
Method: OpenRouter poolside/laguna-s-2.1:free, temperature 0.

Notes:
- CPU-only box: 8 cores, 15 GB RAM, no GPU.
- Incomplete battery
- Provisional.

Results:
completed: 4/8 (code x2, reasoning x2), all passed
rate_limited: 4 tasks got 429 upstream rate limit; a 3rd run hit OpenRouter's free-models-per-day cap
run1.code: passed, 6/6 unit cases, wall 1.99 s, 37 out tokens
run1.reason1: passed, expect 5, wall 13.33 s, 565 out tokens
run1.reason2: passed, expect 5, wall 21.05 s, 703 out tokens
run1.summary: no output, HTTP 429, wall 0.23 s
run2.code: passed, 6/6 unit cases, wall 1.18 s, 38 out tokens
run2.reason1: no output, HTTP 429, wall 0.25 s
run2.reason2: no output, HTTP 429, wall 0.24 s
run2.summary: no output, HTTP 429, wall 0.31 s
run3: all 4 tasks HTTP 429, walls 0.13-0.24 s, no output

Pros:
- Code passed fast (1.2-2.0 s, 37 tokens, no reasoning overhead)
- Both trick questions right

Cons:
- Spent 13-21 s and 565-703 reasoning tokens on simple trick questions
- Upstream rate limits made half the battery fail

Verdict: Every Laguna S 2.1 answer that came back was correct, and its code answers were lean and fast. But half our requests were rate-limited on the free tier, and it overthought simple riddles for up to 21 seconds. The quality looks promising; free availability today was poor. Incomplete run, so the score is provisional.

Score line: 4/4 completed tasks passed, but 4/8 were rate-limited. Provisional.