Public

arduino/arduino-ide

Updated: 8/14/2026

Languages

TypeScript49.8%JavaScript48.4%CSS1.6%HTML0.2%Shell<0.1%
3 Models30 Tasks

Arduino IDE 2.x

Harness

1

Mini-SWE-agent
13 / 30

$0.03

2m15s

2

Mini-SWE-agent
12 / 30

$1.91

16m32s

3

Mini-SWE-agent
10 / 30

$0.64

12m29s

Key Takeaways

  • GPT-5.6 Luna (High) with Mini-SWE-agent scores 43.33%, ahead of Gemini 3.6 Flash with Mini-SWE-agent at 40% and GLM 5.2 (Fireworks) with Mini-SWE-agent at 33.33%.
  • GPT-5.6 Luna (High) with Mini-SWE-agent has the lowest cost per test at $0.03 and shortest latency at 135.11 seconds among the supplied runs.

Model Comparison

Accuracy

43.33%

Gpt 5.6 Luna High

40.00%

Gemini 3.6 Flash

Task outcomes

30 tasks

Both
Gpt 5.6 Luna High only
Gemini 3.6 Flash only
Neither
Not attempted

Cost / test

$0.03

Gpt 5.6 Luna High

$1.91

Gemini 3.6 Flash

Cost distribution

$0.00$2.41$4.83

Latency

2m 15s

Gpt 5.6 Luna High

16m 32s

Gemini 3.6 Flash

Latency distribution

0s19m 21s38m 42s

Cost Analysis

Cost / Test vs. Accuracy
ACCURACYCOST

Average Token Use / Test

Token Usage
InputOutputReasoningCache readCache write
Gemini 3.6 Flash
3.7M
GLM 5.2
2.6M
openai/gpt-5.6-luna-high
606K

Cost is the clearest tradeoff in this comparison. openai/gpt-5.6-luna-high leads at 43.33% for $0.03 per test. No other model in this comparison is cheaper.

Latency Analysis

Latency vs. Accuracy
ACCURACYLATENCY

Average Response Time / Test

Response Time
Gemini 3.6 Flash
16m 32s
GLM 5.2
12m 29s
openai/gpt-5.6-luna-high
2m 15s

openai/gpt-5.6-luna-high is both the most accurate and fastest model in this comparison at 43.33% and 2m 15s.

Tasks with failures

Models
Gpt 5.6 Luna High
Gemini 3.6 Flash
GLM 5.2

Task detail

b4a3a33

Issue statement

Add support for forcing a clean sketch verification/compile. Users must be able to trigger a clean verify with Ctrl/Cmd+Shift+R or by Shift-clicking the Verify toolbar button; the toolbar should show a clean-verify tooltip while Shift is held. The clean choice must be propagated through the compile options to Arduino CLI's compile request so it performs a full recompile. Normal verification must continue to use the non-clean behavior.

View Hidden Tests
diff --git a/arduino-ide-extension/src/test/node/core-service-impl.test.ts b/arduino-ide-extension/src/test/node/core-service-impl.test.tsindex ac419dd2..a07feddb 100644--- a/arduino-ide-extension/src/test/node/core-service-impl.test.ts+++ b/arduino-ide-extension/src/test/node/core-service-impl.test.ts@@ -5,9 +5,39 @@ import {   Port, } from '../../common/protocol/boards-service'; import { Port as RpcPort } from '../../node/cli-protocol/cc/arduino/cli/commands/v1/port_pb';+import { Instance } from '../../node/cli-protocol/cc/arduino/cli/commands/v1/common_pb'; import { CoreServiceImpl } from '../../node/core-service-impl';  describe('core-service-impl', () => {+  describe('compileRequest', () => {+    const options = (clean?: boolean) =>+      ({+        sketch: { uri: 'file:///tmp/CleanCompile/CleanCompile.ino' },+        verbose: false,+        optimizeForDebug: false,+        sourceOverride: {},+        clean,+      } as any);++    it('should request a clean compile when clean is enabled', () => {+      const request = new CoreServiceImpl()['compileRequest'](+        options(true),+        new Instance()+      );++      expect(request.getClean()).to.equal(true);+    });++    it('should not request a clean compile by default', () => {+      const request = new CoreServiceImpl()['compileRequest'](+        options(),+        new Instance()+      );++      expect(request.getClean()).to.equal(false);+    });+  });+   describe('createPort', () => {     it("should map the 'undefined' port object to an 'undefined' gRPC port value", () => {       const actual = new CoreServiceImpl()['createPort'](undefined);