The model the fleet runs
DeepSeek code review, measured
All 8 of IonWarp’s agents start on DeepSeek V4.1 Flash. Below is how it and DeepSeek V4 Pro review, lane by lane. A DeepSeek model is the best model we ran on AI Architecture and CI/CD Performance; on general code review, DeepSeek V4.1 Flash scores F1 49.2% at $0.102 a review.
| Lane | Model | F1 | Rank among models | Time (p50) | $ / review |
|---|---|---|---|---|---|
| Code Review benchmark | DeepSeek V4 Pro | 51.4% | 2 of 5 | 5m 59s | $0.165 |
| Code Review benchmark | DeepSeek V4.1 Flash | 49.2% | 3 of 5 | 6m 56s | $0.102 |
| Security Review benchmark | DeepSeek V4.1 Flash | 62.5% | 2 of 7 | 5m 56s | $0.063 |
| Security Review benchmark | DeepSeek V4 Pro | 34.8% | 6 of 7 | 2m 40s | $0.131 |
| AI Architecture benchmark | DeepSeek V4 Pro | 44.4% | 3 of 8 | 1m 52s | $0.102 |
| AI Architecture benchmark | DeepSeek V4.1 Flash | 47.1% | 1 of 8 | 6m 14s | $0.050 |
| Log & Errors benchmark | DeepSeek V4.1 Flash | 59.0% | 2 of 5 | 4m 08s | $0.039 |
| Docs Review benchmark | DeepSeek V4 Pro | 65.2% | 2 of 3 | 3m 00s | $0.064 |
| API Performance benchmark | DeepSeek V4 Pro | 52.2% | 3 of 6 | 1m 21s | $0.042 |
| API Performance benchmark | DeepSeek V4.1 Flash | 53.3% | 2 of 6 | 6m 27s | $0.043 |
| SOC 2 Review benchmark | DeepSeek V4.1 Flash | 66.7% | 1 of 5 | 5m 56s | $0.038 |
| SOC 2 Review benchmark | DeepSeek V4 Pro | 56.3% | 4 of 5 | 2m 21s | $0.068 |
| CI/CD Performance benchmark | DeepSeek V4 Pro | 72.7% | 1 of 5 | 1m 22s | $0.032 |
| CI/CD Performance benchmark | DeepSeek V4.1 Flash | 59.5% | 2 of 5 | 54s | $0.0095 |
Highlighted rows are what the agent runs today. Green: the best model on that lane.
Code review
DeepSeek against every model and app
| Model / App | Reviewer Score | F1 | Latency | $/run |
|---|---|---|---|---|
| IonWarp · Muse Spark 1.3 | 71.2 | 56.6% | 3m 18s | $0.424 |
| Qodo v2 | 64.7 | 58.5% | 5m 40s | ~$1.67 |
| IonWarp · DeepSeek V4 Pro | 48.7 | 51.4% | 5m 59s | $0.165 |
| IonWarp · DeepSeek V4.1 FlashRuns on your PRs | 41.0 | 49.2% | 6m 56s | $0.102 |
| IonWarp · GLM 5.3 Flash | 40.8 | 42.8% | 2m 15s | $0.018 |
| Greptile v4.1 | 39.9 | 51.1% | 4m 58s | ~$1.75 |
| Cursor Bugbot | 36.2 | 51.4% | 7m 52s | ~$1.25 |
| IonWarp · GPT 5.6 Luna Pro | 31.8 | 41.6% | 1m 47s | $0.116 |
| CodeRabbit | 3.1 | 42.1% | 7m 34s | ~$1.50 |
Reviewer Score = (3 × F1 + time + $/run) ÷ 5, each part 0 to 100 within this table. IonWarp rows cost the model’s inference; ~$ is each app’s list price for 5 seats at 100 reviews a month. Pinned Martian Core reproduction. Release 2026-10-12, CC BY 4.0.
FAQ
DeepSeek for review
Is DeepSeek good for code review?
For some reviews. A DeepSeek model is the best model we ran on AI Architecture (DeepSeek V4.1 Flash, F1 47.1%), CI/CD Performance (DeepSeek V4 Pro, F1 72.7%). On general code review, DeepSeek V4.1 Flash scored F1 49.2% and V4 Pro 51.4%; Muse Spark 1.3 scored 56.6%.
Which DeepSeek model does IonWarp use?
DeepSeek V4.1 Flash, at reasoning effort medium and temperature 0. All 8 of IonWarp’s agents start on DeepSeek V4.1 Flash. An agent moves to another model when IonWarpBench finds one that clears its quality bar for less.
Does my code go to DeepSeek?
No. IonWarp excludes DeepSeek’s own API: the open DeepSeek models run on other hosts through OpenRouter, and only on providers that do not keep or train on what they receive.
How much does a DeepSeek review cost?
DeepSeek V4.1 Flash costs $0.102 in model charges per code review on the benchmark (median 6m 56s). On IonWarp that comes out of the review spend your plan includes.
Did you measure the code DeepSeek writes?
No. We measured DeepSeek as a reviewer of other people’s pull requests. For the code it writes, read a coding benchmark.
Read next
Read next

DeepSeek on every PR, measured every release.
Try for free14 days free, no card. Then Pro is $49 a month per workspace with 5 seats, or Max $149 with 10.