DeepSeek V4 Pro (Workers AI) vs Kimi K2.7 Code (Workers AI)
Kimi K2.7 Code (Workers AI) reads a 26.2K-token pull request for $0.0249, 28% less than DeepSeek V4 Pro (Workers AI) ($0.0346). Both sit in the mid-range price tier, so the choice between them is about review quality more than budget. The benchmark cannot separate their recall: DeepSeek V4 Pro (Workers AI) minus Kimi K2.7 Code (Workers AI) is +16 pts (95% interval +8 pts to +24 pts).
Side by side
| Figure | DeepSeek V4 Pro (Workers AI) | Kimi K2.7 Code (Workers AI) |
|---|---|---|
| Price tier | Mid-range | Mid-range |
| Bugs caughtshare of known bugs | 54% (better) | 38% |
| False alarmsper clean PR | 2.9 | 2.6 (better) |
| Cost per reviewmeasured | $0.0643 | $0.0567 (better) |
| Cost per bug caught | $0.0872 | $0.0805 (better) |
| Exact linecatches on the expected lines | 89% (better) | 79% |
| Review timemean | 390s (better) | 610s |
| Typical review calllist price | $0.0442 | $0.0346 (better) |
| Input / 1M tokens | $1.32 | $0.95 (better) |
| Output / 1M tokens | $3.96 (better) | $4.00 |
| Context window | 1M | 262.1K |
Head to head
The known bugs both were scored on, by who caught each in at least one run.
By language
| Model | All | TypeScript6 bugs | Go4 bugs | Java4 bugs | Rust4 bugs | C#3 bugs | JavaScript3 bugs | Kotlin3 bugs | PHP3 bugs | Python3 bugs | Ruby3 bugs | Swift3 bugs | C++2 bugs | C1 bug |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 Pro (Workers AI) | 54% | 79% | 25% | 49% | 54% | 59% | 17% | 30% | 69% | 72% | 67% | 75% | 38% | 33% |
| Kimi K2.7 Code (Workers AI) | 38% | 75% | 26% | 31% | 46% | 19% | 8% | 14% | 42% | 62% | 33% | 46% | 6% | 40% |
Of the 42 known bugs both were scored on, both caught 30 and neither caught 5. DeepSeek V4 Pro (Workers AI) caught 4 bugs that Kimi K2.7 Code (Workers AI) missed: 2 in Ruby, 1 in C++ and 1 in Java. Kimi K2.7 Code (Workers AI) caught 3 bugs that DeepSeek V4 Pro (Workers AI) missed: 2 in Go and 1 in TypeScript.
On pull requests with no known bug, DeepSeek V4 Pro (Workers AI) raised 2.9 false alarms per review and Kimi K2.7 Code (Workers AI) 2.6. A measured review cost $0.0643 on DeepSeek V4 Pro (Workers AI), 13% more than $0.0567 on Kimi K2.7 Code (Workers AI).
| Pull request | Language · kind | DeepSeek V4 Pro (Workers AI) | Kimi K2.7 Code (Workers AI) |
|---|---|---|---|
| jhy/jsoupStore cached child els in Attributes, not field | Java · Concurrency | 3/6 | 0/16 |
| rails/railsReimplement Action Cable redis adapter with redis-client | Ruby · Logic | 2/4 | 0/5 |
| socketry/asyncBarrier wait in completion order + `Queue#close`. | Ruby · Concurrency | 2/4 | 0/5 |
| ggml-org/llama.cpprpc : cache and reuse compute graphs | C++ · Security | 1/4 | 0/5 |
| colinhacks/zodfeat(v4): parse input containing reference cycles | TypeScript · Resource handling | 0/4 | 1/3 |
| go-gitea/giteaperf(gitdiff): optimize inline diff highlighting using cache | Go · Error handling | 0/3 | 1/4 |
| golang/gocrypto/tls: add LocalCertificate to ConnectionState | Go · Logic | 0/4 | 2/4 |
Defects caught over all runs: a PR reviewed three times with one known bug shows 3/3 when it was caught every time.
Every bug each was scored on: DeepSeek V4 Pro (Workers AI) and Kimi K2.7 Code (Workers AI).