Claude Haiku 5.5 vs DeepSeek V4 Flash (Workers AI)
Reading a 26.2K-token pull request costs $0.0115 on DeepSeek V4 Flash (Workers AI), 4.4× the $0.0026 it costs on Claude Haiku 5.5. Above 100K tokens long-context rates apply, and DeepSeek V4 Flash (Workers AI) becomes the cheaper one. Both sit in the economy price tier, so the choice between them is about review quality more than budget. The benchmark cannot separate their recall: Claude Haiku 5.5 minus DeepSeek V4 Flash (Workers AI) is +15 pts (95% interval +1 pt to +29 pts).
Side by side
| Figure | Claude Haiku 5.5 | DeepSeek V4 Flash (Workers AI) |
|---|---|---|
| Price tier | Economy | Economy |
| Bugs caughtshare of known bugs | 64% (better) | 49% |
| False alarmsper clean PR | 1.4 | 1.3 (better) |
| Cost per reviewmeasured | $0.0038 (better) | $0.0178 |
| Cost per bug caught | $0.0064 (better) | $0.0309 |
| Exact linecatches on the expected lines | 82% | 88% (better) |
| Review timemean | 15s (better) | 137s |
| Typical review calllist price | $0.0038 (better) | $0.0147 |
| Input / 1M tokens | $0.10 (better) | $0.44 |
| Output / 1M tokens | $0.50 (better) | $1.32 |
| Context window | 1M | 1M |
- Above 100K tokens, long-context rates mean DeepSeek V4 Flash (Workers AI) is the cheaper of the two.
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Claude Haiku 5.5 | 64% | 83% | 75% | 40% | 75% | 67% | 33% | 67% | 100% | 39% | 67% | 67% | 50% | 0% |
| DeepSeek V4 Flash (Workers AI) | 49% | 46% | 25% | 55% | 46% | 22% | 56% | 61% | 78% | 61% | 44% | 67% | 50% | 0% |
Of the 42 known bugs both were scored on, both caught 24 and neither caught 8. Claude Haiku 5.5 caught 4 bugs that DeepSeek V4 Flash (Workers AI) missed: 2 in TypeScript, 1 in C# and 1 in Go. DeepSeek V4 Flash (Workers AI) caught 6 bugs that Claude Haiku 5.5 missed: 2 in Java, 1 in JavaScript, 1 in Kotlin, 1 in Python and 1 in Swift.
On pull requests with no known bug, Claude Haiku 5.5 raised 1.4 false alarms per review and DeepSeek V4 Flash (Workers AI) 1.3. A measured review cost $0.0038 on Claude Haiku 5.5, 79% less than $0.0178 on DeepSeek V4 Flash (Workers AI).
| Pull request | Language · kind | Claude Haiku 5.5 | DeepSeek V4 Flash (Workers AI) |
|---|---|---|---|
| OrchardCMS/OrchardCoreSimplify using `StringValues` in Liquid templates | C# · Security | 1/1 | 0/3 |
| colinhacks/zodfeat(v4): parse input containing reference cycles | TypeScript · Resource handling | 1/1 | 0/4 |
| go-gitea/giteaperf(gitdiff): optimize inline diff highlighting using cache | Go · Error handling | 1/1 | 0/3 |
| trpc/trpcfix(server): pass maxDurationMs timeout signal to subscription handlers | TypeScript · Concurrency | 1/1 | 0/1 |
| aio-libs/aiohttpkeep websocket compression state across interleaved control frames | Python · Logic | 0/1 | 2/4 |
| jhy/jsoupStore cached child els in Attributes, not field | Java · Concurrency | 0/2 | 2/6 |
| kotest/kotestFix DataClassEq comparator | Kotlin · Logic | 0/1 | 1/3 |
| mockito/mockitoAdd core API to enable Kotlin singleton mocking | Java · API misuse | 0/1 | 1/4 |
| pinojs/thread-streamrefactor: replace setTimeout with Atomics.waitAsync | JavaScript · Concurrency | 0/1 | 3/3 |
| pointfreeco/swift-navigationUIKit: Fix switch presented child double dismiss | Swift · Logic | 0/1 | 3/3 |
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: Claude Haiku 5.5 and DeepSeek V4 Flash (Workers AI).