Starter
Try AI PR review.
- 2 reviewers
- Code Review
- Security Review
- 3 seats included
- 15,000 credits to start, then 5,000/mo
- About 70 reviews to start, then about 25 a month
- Unlimited repos
Muse Spark 1.3 is the best model we have run on general code review. On 50 open-source pull requests with 158 known issues, it found 79; 42 of its findings matched none: an F1 of 56.6%, at 42 cents.[1] Only the Qodo app scored higher, 58.5%. We ran it on eight of IonWarp's nine review types (how the benchmarks work).
What lands on your pull request · Code Review
billing/checkout.ts:42 · A timed-out response can charge the order twice- await payments.charge(order);+ await payments.charge(order, {+ idempotencyKey: order.id,+ });
IonWarp's Code Review still runs DeepSeek V4.1 Flash by default: it scored 49.2% and costs 10 cents a review, about a quarter of Muse Spark's price.[1] Muse Spark is what a failed Code Review or API Performance review retries on. On most specialist reviews, cheaper models scored higher.
Ahead of DeepSeek V4 Pro (51.4%) and every other model we ran.[1] Best LLM for code review has the full table. In Discourse #8, findMembers() became async and nothing waited for it, so the members list could render empty. Of nine reviewers, only Muse Spark flagged it.[2]
Code ReviewOn Starter, Pro and Max
56.0% on 18 pull requests, in 42 seconds for about 6 cents. Greptile (76.9%), Bugbot (71.4%) and Qodo (66.7%) scored higher.[3]
API PerformanceOn Max
59.5% on 19 pull requests, behind Gemini 3.8 Flash (68.6%) and DeepSeek V4 Pro (65.2%).[4]
Docs ReviewOn Pro and Max
Security 44.4% at 26 cents a review, against 66.7% for GLM 5.3 Flash at about a cent.[5] SOC 2 55.2% and CI/CD 44.4%, last of five models on both;[6][7] logs 43.6%, third of five.[8]
| Review | Model | F1 | Precision | Recall | $ per review | Median time | Rank |
|---|---|---|---|---|---|---|---|
| Code Review50 of 50 PRs | Muse Spark 1.3 | 56.6% | 65.3% | 50.0% | $0.42 | 3m 18s | 1 of 5 |
| Security Review13 of 13 PRs | Muse Spark 1.3 | 44.4% | 66.7% | 33.3% | $0.26 | 2m 30s | 5 of 7 |
| AI Architecture Review11 of 11 PRs | Muse Spark 1.3 | 28.6% | 20.0% | 50.0% | $0.19 | 2m 4s | 7 of 8 |
| Log & Errors Review17 of 17 PRs | Muse Spark 1.3 | 43.6% | 50.0% | 38.7% | $0.08 | 56 sec | 3 of 5 |
| Docs Review19 of 19 PRs | Muse Spark 1.3 | 59.5% | 55.0% | 64.7% | $0.09 | 48 sec | 3 of 3 |
| API Performance Review18 of 18 PRs | Muse Spark 1.3 | 56.0% | 58.3% | 53.8% | $0.06 | 42 sec | 1 of 6 |
| SOC 2 Review20 of 20 PRs | Muse Spark 1.3 | 55.2% | 57.1% | 53.3% | $0.11 | 48 sec | 5 of 5 |
| CI/CD Performance18 of 18 PRs | Muse Spark 1.3 | 44.4% | 42.9% | 46.2% | $0.03 | 49 sec | 5 of 5 |
Rank is among the models we ran on that review, by F1; a tie goes to the cheaper one. See the benchmark
One workspace price with seats included. Credits pool across your team. Unlimited repos on every plan.
Try AI PR review.
Get AI PR review right from day 1.
Scale AI PR review.
Custom reviewers + credits · GitHub Enterprise · SSO/SAML + audit logs · Custom security review + SLA
On general code review, yes: it was the best model we ran, with an F1 of 56.6% on 50 pull requests, behind only the Qodo app. On security, SOC 2, CI/CD and logs, cheaper models scored higher.
Price. On the code benchmark it costs 42 cents a review against 10 for DeepSeek V4.1 Flash, which IonWarp runs for Code Review. A failed Code Review review retries on Muse Spark.
Muse Spark found more known issues, 79 to 64, at 42 cents a review against 16. DeepSeek V4 Pro was more precise: 70% of its findings matched a known issue, against 65% for Muse Spark.
On code review its median was 3 minutes 18 seconds a review. On API performance and docs it took under a minute.