DeepSeek V4 Flash (Workers AI) vs GLM-5.3 (Workers AI)
Reading a 26.2K-token pull request costs $0.0367 on GLM-5.3 (Workers AI), 3.2× the $0.0115 it costs on DeepSeek V4 Flash (Workers AI). DeepSeek V4 Flash (Workers AI) is a economy model and GLM-5.3 (Workers AI) a mid-range one. The benchmark cannot separate their recall: DeepSeek V4 Flash (Workers AI) minus GLM-5.3 (Workers AI) is −12 pts (95% interval −23 pts to −1 pt).
Side by side
| Figure | DeepSeek V4 Flash (Workers AI) | GLM-5.3 (Workers AI) |
|---|---|---|
| Price tier | Economy | Mid-range |
| Bugs caughtshare of known bugs | 49% | 61% (better) |
| False alarmsper clean PR | 1.3 (better) | 1.9 |
| Cost per reviewmeasured | $0.0178 (better) | $0.0311 |
| Cost per bug caught | $0.0309 (better) | $0.0559 |
| Exact linecatches on the expected lines | 88% (better) | 85% |
| Review timemean | 137s | 65s (better) |
| Typical review calllist price | $0.0147 (better) | $0.0473 |
| Input / 1M tokens | $0.44 (better) | $1.40 |
| Output / 1M tokens | $1.32 (better) | $4.40 |
| Context window | 1M | 1M |
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 Flash (Workers AI) | 49% | 46% | 25% | 55% | 46% | 22% | 56% | 61% | 78% | 61% | 44% | 67% | 50% | 0% |
| GLM-5.3 (Workers AI) | 61% | 78% | 19% | 46% | 73% | 48% | 33% | 72% | 44% | 58% | 67% | 100% | 100% | 67% |
Of the 42 known bugs both were scored on, both caught 29 and neither caught 5. DeepSeek V4 Flash (Workers AI) caught one bug that GLM-5.3 (Workers AI) missed: 1 in PHP. GLM-5.3 (Workers AI) caught 7 bugs that DeepSeek V4 Flash (Workers AI) missed: 3 in TypeScript, 1 in C, 1 in C++, 1 in C# and 1 in Rust.
On pull requests with no known bug, DeepSeek V4 Flash (Workers AI) raised 1.3 false alarms per review and GLM-5.3 (Workers AI) 1.9. A measured review cost $0.0178 on DeepSeek V4 Flash (Workers AI), 43% less than $0.0311 on GLM-5.3 (Workers AI).
| Pull request | Language · kind | DeepSeek V4 Flash (Workers AI) | GLM-5.3 (Workers AI) |
|---|---|---|---|
| symfony/symfony[Security] Allow multiple OIDC discovery endpoints | PHP · Security | 1/3 | 0/3 |
| OrchardCMS/OrchardCoreSimplify using `StringValues` in Liquid templates | C# · Security | 0/3 | 2/3 |
| TanStack/queryfix(query-core): fix hydration bugs for already resolved promises | TypeScript · Logic | 0/4 | 3/4 |
| c-ares/c-aresEvent Thread could stall when not notified of new queries on existing connections that are in a bad state | C · Resource handling | 0/3 | 2/3 |
| colinhacks/zodfeat(v4): parse input containing reference cycles | TypeScript · Resource handling | 0/4 | 1/4 |
| ggml-org/llama.cpprpc : cache and reuse compute graphs | C++ · Security | 0/3 | 3/3 |
| nushell/nushellreorganize nushell configuration | Rust · Error handling | 0/4 | 1/4 |
| trpc/trpcfix(server): pass maxDurationMs timeout signal to subscription handlers | TypeScript · Concurrency | 0/1 | 2/3 |
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: DeepSeek V4 Flash (Workers AI) and GLM-5.3 (Workers AI).