Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

cmd: fix context limits for droid and add qwen3-coder-next ctx (#14112)

Parth Sareen42e1d49fbe56
diff --git a/cmd/config/droid.go b/cmd/config/droid.go
index 340cda4fc..b2a0d9693 100644
--- a/cmd/config/droid.go
+++ b/cmd/config/droid.go
@@ -1,6 +1,7 @@
package config
import (
+ "context"
"encoding/json"
"fmt"
"os"
@@ -8,6 +9,7 @@ import (
"path/filepath"
"slices"
+ "github.com/ollama/ollama/api"
"github.com/ollama/ollama/envconfig"
)
@@ -112,9 +114,17 @@ func (d *Droid) Edit(models []string) error {
}
// Build new Ollama model entries with sequential indices (0, 1, 2, ...)
+ client, _ := api.ClientFromEnvironment()
+
var newModels []any
var defaultModelID string
for i, model := range models {
+ maxOutput := 64000
+ if isCloudModel(context.Background(), client, model) {
+ if l, ok := lookupCloudModelLimit(model); ok {
+ maxOutput = l.Output
+ }
+ }
modelID := fmt.Sprintf("custom:%s-%d", model, i)
newModels = append(newModels, modelEntry{
Model: model,
@@ -122,7 +132,7 @@ func (d *Droid) Edit(models []string) error {
BaseURL: envconfig.Host().String() + "/v1",
APIKey: "ollama",
Provider: "generic-chat-completion-api",
- MaxOutputTokens: 64000,
+ MaxOutputTokens: maxOutput,
SupportsImages: false,
ID: modelID,
Index: i,
diff --git a/cmd/config/droid_test.go b/cmd/config/droid_test.go
index 58b4ea9f9..f13c3e936 100644
--- a/cmd/config/droid_test.go
+++ b/cmd/config/droid_test.go
@@ -1251,6 +1251,55 @@ func TestDroidEdit_LargeNumberOfModels(t *testing.T) {
}
}
+func TestDroidEdit_LocalModelDefaultMaxOutput(t *testing.T) {
+ d := &Droid{}
+ tmpDir := t.TempDir()
+ setTestHome(t, tmpDir)
+
+ settingsDir := filepath.Join(tmpDir, ".factory")
+ settingsPath := filepath.Join(settingsDir, "settings.json")
+
+ if err := d.Edit([]string{"llama3.2"}); err != nil {
+ t.Fatal(err)
+ }
+
+ data, _ := os.ReadFile(settingsPath)
+ var settings map[string]any
+ json.Unmarshal(data, &settings)
+
+ models := settings["customModels"].([]any)
+ entry := models[0].(map[string]any)
+ if entry["maxOutputTokens"] != float64(64000) {
+ t.Errorf("local model maxOutputTokens = %v, want 64000", entry["maxOutputTokens"])
+ }
+}
+
+func TestDroidEdit_CloudModelLimitsUsed(t *testing.T) {
+ // Verify that every cloud model in cloudModelLimits has a valid output
+ // value that would be used for maxOutputTokens when isCloudModel returns true.
+ // :cloud suffix stripping must also work since that's how users specify them.
+ for name, expected := range cloudModelLimits {
+ t.Run(name, func(t *testing.T) {
+ l, ok := lookupCloudModelLimit(name)
+ if !ok {
+ t.Fatalf("lookupCloudModelLimit(%q) returned false", name)
+ }
+ if l.Output != expected.Output {
+ t.Errorf("output = %d, want %d", l.Output, expected.Output)
+ }
+ // Also verify :cloud suffix lookup
+ cloudName := name + ":cloud"
+ l2, ok := lookupCloudModelLimit(cloudName)
+ if !ok {
+ t.Fatalf("lookupCloudModelLimit(%q) returned false", cloudName)
+ }
+ if l2.Output != expected.Output {
+ t.Errorf(":cloud output = %d, want %d", l2.Output, expected.Output)
+ }
+ })
+ }
+}
+
func TestDroidEdit_ArraysWithMixedTypes(t *testing.T) {
d := &Droid{}
tmpDir := t.TempDir()
diff --git a/cmd/config/opencode.go b/cmd/config/opencode.go
index 59a6f0119..0fe9f8a8f 100644
--- a/cmd/config/opencode.go
+++ b/cmd/config/opencode.go
@@ -39,6 +39,7 @@ var cloudModelLimits = map[string]cloudModelLimit{
"kimi-k2-thinking": {Context: 262_144, Output: 262_144},
"nemotron-3-nano:30b": {Context: 1_048_576, Output: 131_072},
"qwen3-coder:480b": {Context: 262_144, Output: 65_536},
+ "qwen3-coder-next": {Context: 262_144, Output: 32_768},
"qwen3-next:80b": {Context: 262_144, Output: 32_768},
}
diff --git a/cmd/config/opencode_test.go b/cmd/config/opencode_test.go
index 9dc85dd10..8de174588 100644
--- a/cmd/config/opencode_test.go
+++ b/cmd/config/opencode_test.go
@@ -633,6 +633,7 @@ func TestLookupCloudModelLimit(t *testing.T) {
{"deepseek-v3.2", true, 163_840, 65_536},
{"deepseek-v3.2:cloud", true, 163_840, 65_536},
{"qwen3-coder:480b", true, 262_144, 65_536},
+ {"qwen3-coder-next:cloud", true, 262_144, 32_768},
{"llama3.2", false, 0, 0},
{"unknown-model:cloud", false, 0, 0},
}