Gemma 4 26B A4B (Workers AI) vs GPT-6 Luna
Gemma 4 26B A4B (Workers AI) and GPT-6 Luna charge practically the same to read a 26.2K-token pull request ($0.0026). Above 272K tokens long-context rates apply, and Gemma 4 26B A4B (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. In the benchmark GPT-6 Luna caught 41 percentage points more of the known defects than Gemma 4 26B A4B (Workers AI) (95% interval −56 pts to −26 pts, a − b).
Side by side
| Figure | Gemma 4 26B A4B (Workers AI) | GPT-6 Luna |
|---|---|---|
| Price tier | Economy | Economy |
| Bugs caughtshare of known bugs | 26% | 67% (better) |
| False alarmsper clean PR | 0.4 (better) | 0.7 |
| Cost per reviewmeasured | $0.0048 | $0.0028 (better) |
| Cost per bug caught | $0.0209 | $0.0046 (better) |
| Exact linecatches on the expected lines | 91% | 97% (better) |
| Review timemean | 308s | 29s (better) |
| Typical review calllist price | $0.0033 (better) | $0.0038 |
| Input / 1M tokens | $0.10 | $0.10 |
| Output / 1M tokens | $0.30 (better) | $0.50 |
| Context window | 256K | 1.1M |
- Above 272K tokens, long-context rates mean Gemma 4 26B A4B (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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gemma 4 26B A4B (Workers AI) | 26% | 33% | 0% | 30% | 38% | 7% | 0% | 56% | 33% | 44% | 67% | 0% | 0% | 0% |
| GPT-6 Luna | 67% | 83% | 25% | 65% | 75% | 56% | 67% | 67% | 67% | 100% | 67% | 100% | 50% | 0% |
Of the 42 known bugs both were scored on, both caught 14 and neither caught 11. Gemma 4 26B A4B (Workers AI) caught 2 bugs that GPT-6 Luna missed: 1 in Kotlin and 1 in TypeScript, all of them logic bugs. GPT-6 Luna caught 15 bugs that Gemma 4 26B A4B (Workers AI) missed: 3 in Swift, 3 in TypeScript, 2 in JavaScript, 1 in C++, 1 in C#, 1 in Go, 1 in Java, 1 in PHP, 1 in Python and 1 in Rust.
On pull requests with no known bug, Gemma 4 26B A4B (Workers AI) raised 0.4 false alarms per review and GPT-6 Luna 0.7. A measured review cost $0.0048 on Gemma 4 26B A4B (Workers AI), 68% more than $0.0028 on GPT-6 Luna.
| Pull request | Language · kind | Gemma 4 26B A4B (Workers AI) | GPT-6 Luna |
|---|---|---|---|
| withastro/astroPort Go's csrf protection | TypeScript · Logic | 1/2 | 0/1 |
| kotest/kotestFix DataClassEq comparator | Kotlin · Logic | 1/3 | 0/1 |
| OrchardCMS/OrchardCoreSimplify using `StringValues` in Liquid templates | C# · Security | 0/3 | 1/1 |
| TanStack/queryfix(query-core): fix hydration bugs for already resolved promises | TypeScript · Logic | 0/2 | 1/1 |
| aio-libs/aiohttpkeep websocket compression state across interleaved control frames | Python · Logic | 0/3 | 1/1 |
| better-auth/better-authfix(saml): IdP-Initiated Callback Routing | TypeScript · Security | 0/2 | 2/2 |
| colinhacks/zodfeat(v4): parse input containing reference cycles | TypeScript · Resource handling | 0/2 | 1/1 |
| doctrine/ormFix collection filtering API for IN/NOT IN comparisons that require type conversions | PHP · Logic | 0/3 | 1/1 |
| duckdb/duckdbFix vsize = 2 issue in JSON create | C++ · Concurrency | 0/3 | 1/1 |
| go-chi/chifeat: middleware.ClientIP, a replacement for middleware.RealIP | Go · Security | 0/3 | 1/1 |
| groue/GRDB.swiftDatabaseQueue and DatabaseSnapshot reads can access task locals | Swift · Concurrency | 0/2 | 1/1 |
| mockito/mockitoAdd core API to enable Kotlin singleton mocking | Java · API misuse | 0/3 | 1/1 |
| nodejs/undicifeat: add configurable maxPayloadSize for WebSocket | JavaScript · Security | 0/3 | 1/1 |
| pinojs/thread-streamrefactor: replace setTimeout with Atomics.waitAsync | JavaScript · Concurrency | 0/4 | 1/1 |
| pointfreeco/swift-composable-architectureRemove child store from parent when invalidated | Swift · Resource handling | 0/4 | 1/1 |
| pointfreeco/swift-navigationUIKit: Fix switch presented child double dismiss | Swift · Logic | 0/4 | 1/1 |
| tokio-rs/tokiotime: move lazy-registration state into `Sleep` | Rust · Logic | 0/2 | 1/1 |
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: Gemma 4 26B A4B (Workers AI) and GPT-6 Luna.