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
DeepSeek code review is at its best on CI/CD: V4 Pro found 12 of 13 known defects on 18 pull requests, for about 3 cents a review, the best of the five models we ran.[1] On general code review it came second of five models (F1 51.4%); V4.1 Flash, which IonWarp's Code Review runs, came third (49.2%).[2]
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,+ });
On CI/CD, V4 Pro saw a Reanimated workflow key its cache on its own trigger files, so it never hit; V4.1 Flash, which CI/CD Performance runs, missed it.[3] Run a reviewer yourself on your own key and the model fees are your bill; free AI code review compares that route with the hosted free plans.
V4.1 Flash found 11 of the 13 for about a cent a review.[1]
CI/CD PerformanceOn Max
Both trail Muse Spark 1.3 on F1. More of V4 Pro's findings matched a known issue, while V4.1 Flash caught more of the known issues and reported more that matched nothing. The table below has each figure.[2]
Code ReviewOn Starter, Pro and Max
13 of 15 known control gaps found on 20 pull requests, an F1 of 66.7%, tied with GLM 5.3 Flash, which took 34 seconds a review to V4.1 Flash's six minutes.[4]
SOC 2 ReviewOn Max
On 13 security pull requests V4.1 Flash scored 62.5% at about 6 cents a review, and GLM 5.3 Flash 66.7% at about a cent. DeepSeek V4 Pro found only 4 of the 18 known issues.[5]
Security ReviewOn Starter, Pro and Max
DeepSeek V4 Pro found the most stale docs, 15 of 17, but about half its findings matched no known issue. Gemini 3.8 Flash had the top F1, 68.6% to 65.2%.[6]
Docs ReviewOn Pro and Max
The shortest route is a GitHub Action that sends each diff to DeepSeek and comments with the reply. deepseek-review does this, with your key in a repository secret.[7] DeepSeek's API accepts the OpenAI and Anthropic formats, so a script you already have needs a new base URL, key and model name.[8]
DeepSeek's API now calls V4.1 Flash deepseek-flash,[8] and DeepSeek upgraded that model on October 17.[9] The older name deepseek-v4-flash, still the default in deepseek-review,[7] is accepted and answered by the upgraded build.[8] Our V4.1 Flash runs ended by October 6, so the build you reach today is not the one in our tables.[2]
Code Review runs DeepSeek V4.1 Flash by default and leaves one comment on the PR, with the file and line it points to and a suggested fix. The plan includes the model cost, so you hold no DeepSeek key.
Code ReviewOn Starter, Pro and Max
Every score on this page is DeepSeek running as an IonWarp reviewer on our code review benchmarks, and each model reviewed each pull request once.[10] A prompt that holds only the diff is a different reviewer, and we have not measured one. Run yours on a few merged pull requests where you know what went wrong, before you trust it.
| Review | Model | F1 | Precision | Recall | $ per review | Median time | Rank |
|---|---|---|---|---|---|---|---|
| Code Review50 of 50 PRs | DeepSeek V4 Pro | 51.4% | 70.3% | 40.5% | $0.16 | 5m 59s | 2 of 5 |
| Code Review50 of 50 PRs | DeepSeek V4.1 Flash | 49.2% | 49.7% | 48.7% | $0.10 | 6m 56s | 3 of 5 |
| Security Review13 of 13 PRs | DeepSeek V4.1 Flash | 62.5% | 71.4% | 55.6% | $0.06 | 5m 56s | 2 of 7 |
| Security Review13 of 13 PRs | DeepSeek V4 Pro | 34.8% | 80.0% | 22.2% | $0.13 | 2m 40s | 6 of 7 |
| AI Architecture Review11 of 11 PRs | DeepSeek V4.1 Flash | 47.1% | 36.4% | 66.7% | $0.05 | 6m 14s | 1 of 8 |
| AI Architecture Review11 of 11 PRs | DeepSeek V4 Pro | 44.4% | 33.3% | 66.7% | $0.10 | 1m 52s | 3 of 8 |
| Log & Errors Review17 of 17 PRs | DeepSeek V4.1 Flash | 59.0% | 60.0% | 58.1% | $0.04 | 4m 8s | 2 of 5 |
| Docs Review19 of 19 PRs | DeepSeek V4 Pro | 65.2% | 51.7% | 88.2% | $0.06 | 3m 0s | 2 of 3 |
| API Performance Review18 of 18 PRs | DeepSeek V4.1 Flash | 53.3% | 47.1% | 61.5% | $0.04 | 6m 27s | 2 of 6 |
| API Performance Review18 of 18 PRs | DeepSeek V4 Pro | 52.2% | 60.0% | 46.2% | $0.04 | 1m 21s | 3 of 6 |
| SOC 2 Review20 of 20 PRs | DeepSeek V4.1 Flash | 66.7% | 54.2% | 86.7% | $0.04 | 5m 56s | 2 of 5 |
| SOC 2 Review20 of 20 PRs | DeepSeek V4 Pro | 56.3% | 52.9% | 60.0% | $0.07 | 2m 21s | 4 of 5 |
| CI/CD Performance18 of 18 PRs | DeepSeek V4 Pro | 72.7% | 60.0% | 92.3% | $0.03 | 1m 22s | 1 of 5 |
| CI/CD Performance18 of 18 PRs | DeepSeek V4.1 Flash | 59.5% | 45.8% | 84.6% | $0.01 | 54 sec | 2 of 5 |
| Cost Review20 of 20 PRs | DeepSeek V4.1 Flash | 57.1% | 47.6% | 71.4% | $0.07 | 10m 39s | 1 of 3 |
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
For some reviews. DeepSeek V4 Pro had the best result of any model on CI/CD pull requests, and V4.1 Flash tied for first on SOC 2. On general code review Muse Spark 1.3 scored higher, and on security GLM 5.3 Flash scored higher for less.
We measured DeepSeek as a reviewer of other people's code, not the quality of the code it writes. As a reviewer, V4 Pro found 12 of 13 known CI/CD defects, and on general code review it scored an F1 of 51.4%, second of the five models we ran. For the code it writes, read a coding benchmark.
No. The published tables have DeepSeek V4.1 Flash and DeepSeek V4 Pro, and no DeepSeek V4 Flash row. DeepSeek has retired V4 Flash, and its API now answers that name with V4.1 Flash (DeepSeek API docs, Nov 2026).
V4 Pro scored higher on our code benchmark (F1 51.4% to 49.2%) with fewer findings that matched no known issue, at about 16 cents a review against 10. IonWarp still runs V4.1 Flash for Code Review: on reviews of our own repositories from 1 to 21 September 2026 it filed as many P0 and P1 findings per review as V4 Pro (14.8 per 100 on 549 reviews, against 13.5 on 104), for about a quarter of the cost.
We have not benchmarked Claude models as reviewers. Claude Opus 4.5 is the grader on the code review benchmark: it decides whether each finding matches a known issue.
In model fees on our benchmarks, V4.1 Flash cost about 10 cents a code review and 1 to 7 cents on the smaller specialist reviews; V4 Pro cost about 16 cents a code review. IonWarp plans include the model cost: Pro is $49/mo for the workspace.