CodePanel

Gemini 3.8 Flash vs Gemma 4 26B A4B (Workers AI)

Reading a 26.2K-token pull request costs $0.0197 on Gemini 3.8 Flash, 7.5× the $0.0026 it costs on Gemma 4 26B A4B (Workers AI). Gemini 3.8 Flash is a mid-range model and Gemma 4 26B A4B (Workers AI) a economy one. In the benchmark Gemini 3.8 Flash caught 42 percentage points more of the known defects than Gemma 4 26B A4B (Workers AI) (95% interval +25 pts to +58 pts, a − b).

Side by side

Gemini 3.8 Flash and Gemma 4 26B A4B (Workers AI) side by side
FigureGemini 3.8 FlashGemma 4 26B A4B (Workers AI)
Price tierMid-rangeEconomy
Bugs caughtshare of known bugs68% (better)26%
False alarmsper clean PR1.00.4 (better)
Cost per reviewmeasured$0.0862$0.0048 (better)
Cost per bug caught$0.1166$0.0209 (better)
Exact linecatches on the expected lines94% (better)91%
Review timemean427s308s (better)
Typical review calllist price$0.0287$0.0033 (better)
Input / 1M tokens$0.75$0.10 (better)
Output / 1M tokens$3.75$0.30 (better)
Context window1M256K

Head to head

The known bugs both were scored on, by who caught each in at least one run.

Of 42 known bugs both were scored on: Both 13Only Gemini 3.8 Flash 18Only Gemma 4 26B A4B (Workers AI) 3Neither 8

By language

Share of known bugs each model caught, by language
ModelAllTypeScript6 bugsGo4 bugsJava4 bugsRust4 bugsC#3 bugsJavaScript3 bugsKotlin3 bugsPHP3 bugsPython3 bugsRuby3 bugsSwift3 bugsC++2 bugsC1 bug
Gemini 3.8 Flash68%83%75%43%75%78%67%0%50%83%100%100%50%50%
Gemma 4 26B A4B (Workers AI)26%33%0%30%38%7%0%56%33%44%67%0%0%0%

Where they differ

Of the 42 known bugs both were scored on, both caught 13 and neither caught 8. Gemini 3.8 Flash caught 18 bugs that Gemma 4 26B A4B (Workers AI) missed: 3 in Go, 3 in Swift, 2 in C#, 2 in JavaScript, 2 in TypeScript, 1 in C, 1 in C++, 1 in PHP, 1 in Python, 1 in Ruby and 1 in Rust. Gemma 4 26B A4B (Workers AI) caught 3 bugs that Gemini 3.8 Flash missed: 3 in Kotlin, most of them logic bugs.

On pull requests with no known bug, Gemini 3.8 Flash raised 1 false alarms per review and Gemma 4 26B A4B (Workers AI) 0.4. A measured review cost $0.0862 on Gemini 3.8 Flash, 18× the cost of $0.0048 on Gemma 4 26B A4B (Workers AI).

Each known bug, and how often each model caught it
Pull requestLanguage · kindGemini 3.8 FlashGemma 4 26B A4B (Workers AI)
OrchardCMS/OrchardCoreSimplify using `StringValues` in Liquid templatesC# · Security1/10/3
TanStack/queryfix(query-core): fix hydration bugs for already resolved promisesTypeScript · Logic1/10/2
aio-libs/aiohttpkeep websocket compression state across interleaved control framesPython · Logic2/20/3
better-auth/better-authfix(saml): IdP-Initiated Callback RoutingTypeScript · Security1/10/2
doctrine/ormFix collection filtering API for IN/NOT IN comparisons that require type conversionsPHP · Logic2/20/3
dotnet/orleansfix(serialization): cap codec allocationsC# · Logic1/10/4
duckdb/duckdbFix vsize = 2 issue in JSON createC++ · Concurrency1/10/3
go-chi/chifeat: middleware.ClientIP, a replacement for middleware.RealIPGo · Security1/10/3
golang/gocrypto/tls: add LocalCertificate to ConnectionStateGo · Logic2/20/3
groue/GRDB.swiftDatabaseQueue and DatabaseSnapshot reads can access task localsSwift · Concurrency2/20/2
pinojs/thread-streamrefactor: replace setTimeout with Atomics.waitAsyncJavaScript · Concurrency2/20/4
pointfreeco/swift-composable-architectureRemove child store from parent when invalidatedSwift · Resource handling2/20/4
pointfreeco/swift-navigationUIKit: Fix switch presented child double dismissSwift · Logic2/20/4
sindresorhus/p-throttleAdd `weight` optionJavaScript · Logic1/10/3
socketry/asyncBarrier wait in completion order + `Queue#close`.Ruby · Concurrency1/10/3
tokio-rs/tokiotime: move lazy-registration state into `Sleep`Rust · Logic1/10/2
traefik/traefikAdd the option to define custom cipher suites for backend serversTransportGo · API misuse1/10/3
c-ares/c-aresEvent Thread could stall when not notified of new queries on existing connections that are in a bad stateC · Resource handling1/20/4
kotest/kotestFix DataClassEq comparatorKotlin · Logic0/11/3
kotest/kotestRefactor eventually to ensure wall-clock time usageKotlin · Error handling0/13/3
ktorio/ktorKTOR-9171 Redesign ByteReadChannel.readUTF8Line APIKotlin · Logic0/11/3

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: Gemini 3.8 Flash and Gemma 4 26B A4B (Workers AI).