1package provider
2
3import (
4 "context"
5 "math"
6 "net/http"
7 "net/http/httptest"
8 "strings"
9 "testing"
10)
11
12func TestSSEParse(t *testing.T) {
13 r := strings.NewReader("event: foo\ndata: {\"a\":1}\n\ndata: hello\ndata: world\n\n")
14 ch := make(chan sseEvent, 4)
15 go readSSE(r, ch)
16
17 e := <-ch
18 if e.Event != "foo" || e.Data != `{"a":1}` {
19 t.Fatalf("event 1: %+v", e)
20 }
21 e = <-ch
22 if e.Event != "" || e.Data != "hello\nworld" {
23 t.Fatalf("event 2: %+v", e)
24 }
25 if _, ok := <-ch; ok {
26 t.Fatalf("channel not closed")
27 }
28}
29
30func TestModelCatalog(t *testing.T) {
31 if len(Catalog) == 0 {
32 t.Fatal("empty catalog")
33 }
34 if _, err := FindModel("anthropic", "claude-sonnet-4-5"); err != nil {
35 t.Fatal(err)
36 }
37 if _, err := FindModel("openai", "gpt-5"); err != nil {
38 t.Fatal(err)
39 }
40 if _, err := FindModel("", "nope"); err == nil {
41 t.Fatal("expected error")
42 }
43}
44
45func TestVertexGeminiFlashCatalogLimits(t *testing.T) {
46 for _, id := range []string{"gemini-3.1-flash-lite", "gemini-3.5-flash"} {
47 m, err := FindModel("google-vertex", id)
48 if err != nil {
49 t.Fatal(err)
50 }
51 if m.ContextWindow != 1048576 || m.MaxOutput != 65535 || !m.Reasoning {
52 t.Errorf("%s metadata = %+v", id, m)
53 }
54 }
55}
56
57func TestComputeCost(t *testing.T) {
58 m, _ := FindModel("anthropic", "claude-sonnet-4-5")
59 cost := ComputeCost(m, Usage{InputTokens: 1_000_000, OutputTokens: 1_000_000})
60 want := m.PriceInput + m.PriceOutput
61 if cost != want {
62 t.Fatalf("cost=%v want=%v", cost, want)
63 }
64}
65
66func TestComputeCostInputTier(t *testing.T) {
67 m := Model{
68 PriceInput: 1, PriceOutput: 2, PriceCacheRead: 0.1,
69 PriceTierInputTokens: 100,
70 PriceInputAbove: 3, PriceOutputAbove: 4, PriceCacheReadAbove: 0.2,
71 }
72 below := ComputeCost(m, Usage{InputTokens: 50, OutputTokens: 10, CacheReadTokens: 50})
73 wantBelow := (50.0*1 + 10*2 + 50*0.1) / 1_000_000
74 if math.Abs(below-wantBelow) > 1e-12 {
75 t.Fatalf("below-tier cost=%v want=%v", below, wantBelow)
76 }
77
78 above := ComputeCost(m, Usage{InputTokens: 51, OutputTokens: 10, CacheReadTokens: 50})
79 wantAbove := (51.0*3 + 10*4 + 50*0.2) / 1_000_000
80 if math.Abs(above-wantAbove) > 1e-12 {
81 t.Fatalf("above-tier cost=%v want=%v", above, wantAbove)
82 }
83}
84
85func TestAnthropicErrorStatus(t *testing.T) {
86 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
87 w.WriteHeader(http.StatusUnauthorized)
88 _, _ = w.Write([]byte(`{"error":{"type":"auth","message":"bad key"}}`))
89 }))
90 defer srv.Close()
91
92 c := NewAnthropic("x", srv.URL)
93 _, err := c.Stream(context.Background(), Request{Model: "claude-sonnet-4-5"})
94 if err == nil || !strings.Contains(err.Error(), "401") {
95 t.Fatalf("want 401 err, got %v", err)
96 }
97}
98
99func TestAnthropicOAuthUsesCurrentClaudeCodeVersion(t *testing.T) {
100 var userAgent string
101 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
102 userAgent = r.Header.Get("user-agent")
103 w.WriteHeader(http.StatusBadRequest)
104 }))
105 defer srv.Close()
106
107 c := NewAnthropicOAuth("test-token", srv.URL)
108 _, err := c.Stream(context.Background(), Request{Model: "claude-fable-5-1"})
109 if err == nil {
110 t.Fatal("expected test server error")
111 }
112 if want := "claude-cli/2.1.258"; userAgent != want {
113 t.Fatalf("user-agent = %q, want %q", userAgent, want)
114 }
115}
116
117func TestOpenAIErrorStatus(t *testing.T) {
118 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
119 w.WriteHeader(http.StatusBadRequest)
120 _, _ = w.Write([]byte(`{"error":{"message":"nope"}}`))
121 }))
122 defer srv.Close()
123
124 c := NewOpenAI("x", srv.URL)
125 _, err := c.Stream(context.Background(), Request{Model: "gpt-5"})
126 if err == nil || !strings.Contains(err.Error(), "400") {
127 t.Fatalf("want 400 err, got %v", err)
128 }
129}
130
131func TestMiniMaxM3CatalogAndRequest(t *testing.T) {
132 for _, provider := range []string{"minimax", "minimax-cn"} {
133 m, err := FindModel(provider, "MiniMax-M3")
134 if err != nil {
135 t.Fatal(err)
136 }
137 if m.ContextWindow != 1000000 || m.MaxOutput != 131072 || !m.Reasoning || !m.AdaptiveThinkingCompat {
138 t.Errorf("%s metadata = %+v", provider, m)
139 }
140 if m.PriceInput != 0.3 || m.PriceOutput != 1.2 || m.PriceCacheRead != 0.06 || m.PriceTierInputTokens != 512000 || m.PriceInputAbove != 0.6 || m.PriceOutputAbove != 2.4 || m.PriceCacheReadAbove != 0.12 {
141 t.Errorf("%s pricing = %+v", provider, m)
142 }
143 }
144
145 c := NewAnthropicCompat("minimax", "x", "https://api.minimax.io/anthropic").(*anthropicClient)
146 temp := float32(0.7)
147 wire, err := c.buildRequest(Request{
148 Model: "MiniMax-M3",
149 Reasoning: "high",
150 Temperature: &temp,
151 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
152 })
153 if err != nil {
154 t.Fatal(err)
155 }
156 if wire.Thinking == nil || wire.Thinking.Type != "adaptive" || wire.Thinking.BudgetTokens != 0 {
157 t.Fatalf("want adaptive thinking without a budget, got %+v", wire.Thinking)
158 }
159 if wire.OutputConfig != nil {
160 t.Fatalf("MiniMax-M3 must not send output_config, got %+v", wire.OutputConfig)
161 }
162 if wire.Temperature == nil || *wire.Temperature != temp {
163 t.Fatalf("MiniMax-M3 should keep temperature, got %v", wire.Temperature)
164 }
165}
166
167func TestAnthropicAdaptiveThinking(t *testing.T) {
168 c := NewAnthropic("x", "").(*anthropicClient)
169 temp := float32(0.7)
170
171 // Opus 5 -> adaptive thinking, effort set, no budget, no temperature.
172 wire, err := c.buildRequest(Request{
173 Model: "claude-opus-5",
174 Reasoning: "high",
175 Temperature: &temp,
176 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
177 })
178 if err != nil {
179 t.Fatal(err)
180 }
181 if wire.Thinking == nil || wire.Thinking.Type != "adaptive" {
182 t.Fatalf("want adaptive thinking, got %+v", wire.Thinking)
183 }
184 if wire.Thinking.BudgetTokens != 0 {
185 t.Fatalf("adaptive must not send budget_tokens, got %d", wire.Thinking.BudgetTokens)
186 }
187 if wire.OutputConfig == nil || wire.OutputConfig.Effort != "high" {
188 t.Fatalf("want effort=high, got %+v", wire.OutputConfig)
189 }
190 if wire.Temperature != nil {
191 t.Fatalf("adaptive must drop temperature, got %v", *wire.Temperature)
192 }
193
194 // maximum -> xhigh effort.
195 wire, err = c.buildRequest(Request{
196 Model: "claude-opus-5",
197 Reasoning: "maximum",
198 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
199 })
200 if err != nil {
201 t.Fatal(err)
202 }
203 if wire.OutputConfig == nil || wire.OutputConfig.Effort != "xhigh" {
204 t.Fatalf("want effort=xhigh, got %+v", wire.OutputConfig)
205 }
206
207 // max is a separate native tier above xhigh on adaptive models.
208 wire, err = c.buildRequest(Request{
209 Model: "claude-opus-5",
210 Reasoning: "max",
211 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
212 })
213 if err != nil {
214 t.Fatal(err)
215 }
216 if wire.OutputConfig == nil || wire.OutputConfig.Effort != "max" {
217 t.Fatalf("want effort=max, got %+v", wire.OutputConfig)
218 }
219
220 // Sonnet 5 -> adaptive thinking, effort set, no budget, no temperature.
221 wire, err = c.buildRequest(Request{
222 Model: "claude-sonnet-5",
223 Reasoning: "high",
224 Temperature: &temp,
225 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
226 })
227 if err != nil {
228 t.Fatal(err)
229 }
230 if wire.Thinking == nil || wire.Thinking.Type != "adaptive" {
231 t.Fatalf("want adaptive thinking for Sonnet 5, got %+v", wire.Thinking)
232 }
233 if wire.Thinking.BudgetTokens != 0 {
234 t.Fatalf("Sonnet 5 adaptive must not send budget_tokens, got %d", wire.Thinking.BudgetTokens)
235 }
236 if wire.OutputConfig == nil || wire.OutputConfig.Effort != "high" {
237 t.Fatalf("want Sonnet 5 effort=high, got %+v", wire.OutputConfig)
238 }
239 if wire.Temperature != nil {
240 t.Fatalf("Sonnet 5 adaptive must drop temperature, got %v", *wire.Temperature)
241 }
242
243 // Opus 4.5 -> budget-based thinking, no output_config, temperature kept.
244 wire, err = c.buildRequest(Request{
245 Model: "claude-opus-4-5",
246 Reasoning: "high",
247 Temperature: &temp,
248 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
249 })
250 if err != nil {
251 t.Fatal(err)
252 }
253 if wire.Thinking == nil || wire.Thinking.Type != "enabled" || wire.Thinking.BudgetTokens <= 0 {
254 t.Fatalf("want budget thinking, got %+v", wire.Thinking)
255 }
256 if wire.OutputConfig != nil {
257 t.Fatalf("budget models must not send output_config, got %+v", wire.OutputConfig)
258 }
259 if wire.Temperature == nil || *wire.Temperature != temp {
260 t.Fatalf("budget model should keep temperature, got %v", wire.Temperature)
261 }
262}
263
264func TestAnthropicBuildRequestStripsAssistantImages(t *testing.T) {
265 c := NewAnthropic("x", "").(*anthropicClient)
266 wire, err := c.buildRequest(Request{
267 Model: "claude-sonnet-4-5",
268 Messages: []Message{
269 {Role: RoleUser, Content: []Content{TextBlock{Text: "make an image"}}},
270 {Role: RoleAssistant, Content: []Content{
271 TextBlock{Text: "done"},
272 ImageBlock{MimeType: "image/png", Data: []byte("png")},
273 TextBlock{Text: "Saved image: `zot-gemini-image-x.png`"},
274 }},
275 {Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
276 },
277 })
278 if err != nil {
279 t.Fatal(err)
280 }
281 if len(wire.Messages) != 3 {
282 t.Fatalf("messages=%d", len(wire.Messages))
283 }
284 assistant := wire.Messages[1]
285 if assistant.Role != "assistant" {
286 t.Fatalf("role=%q", assistant.Role)
287 }
288 if len(assistant.Content) != 2 {
289 t.Fatalf("assistant content=%+v", assistant.Content)
290 }
291 for _, b := range assistant.Content {
292 if _, ok := b.(anthImageBlock); ok {
293 t.Fatalf("assistant image block was not stripped: %+v", assistant.Content)
294 }
295 }
296}
297
298func TestAnthropicBuildRequestDeduplicatesToolResults(t *testing.T) {
299 c := NewAnthropic("x", "").(*anthropicClient)
300 wire, err := c.buildRequest(Request{
301 Model: "claude-sonnet-4-5",
302 Messages: []Message{
303 {Role: RoleUser, Content: []Content{TextBlock{Text: "read the file"}}},
304 {Role: RoleAssistant, Content: []Content{
305 ToolCallBlock{ID: "tool-1", Name: "read"},
306 }},
307 {Role: RoleTool, Content: []Content{
308 ToolResultBlock{CallID: "tool-1", Content: []Content{TextBlock{Text: "first result"}}},
309 }},
310 {Role: RoleTool, Content: []Content{
311 ToolResultBlock{CallID: "tool-1", Content: []Content{TextBlock{Text: "duplicate result"}}},
312 }},
313 },
314 })
315 if err != nil {
316 t.Fatal(err)
317 }
318 if len(wire.Messages) != 3 {
319 t.Fatalf("messages=%d, want 3 after dropping duplicate result message", len(wire.Messages))
320 }
321 blocks := wire.Messages[2].Content
322 if len(blocks) != 1 {
323 t.Fatalf("tool result blocks=%d, want 1", len(blocks))
324 }
325 result, ok := blocks[0].(anthToolResultBlock)
326 if !ok {
327 t.Fatalf("tool result block type=%T", blocks[0])
328 }
329 if string(result.Content) != `"first result"` {
330 t.Fatalf("tool result content=%s", result.Content)
331 }
332}
333
334func TestAnthropicStreamHappyPath(t *testing.T) {
335 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
336 w.Header().Set("content-type", "text/event-stream")
337 w.WriteHeader(http.StatusOK)
338 fl, _ := w.(http.Flusher)
339 write := func(s string) {
340 _, _ = w.Write([]byte(s))
341 if fl != nil {
342 fl.Flush()
343 }
344 }
345 write("event: message_start\ndata: {\"message\":{\"usage\":{\"input_tokens\":10,\"output_tokens\":0}}}\n\n")
346 write("event: content_block_start\ndata: {\"index\":0,\"content_block\":{\"type\":\"text\",\"text\":\"\"}}\n\n")
347 write("event: content_block_delta\ndata: {\"index\":0,\"delta\":{\"type\":\"text_delta\",\"text\":\"hi\"}}\n\n")
348 write("event: content_block_stop\ndata: {\"index\":0}\n\n")
349 write("event: message_delta\ndata: {\"delta\":{\"stop_reason\":\"end_turn\"},\"usage\":{\"output_tokens\":5}}\n\n")
350 write("event: message_stop\ndata: {}\n\n")
351 }))
352 defer srv.Close()
353
354 c := NewAnthropic("x", srv.URL)
355 evs, err := c.Stream(context.Background(), Request{Model: "claude-sonnet-4-5"})
356 if err != nil {
357 t.Fatal(err)
358 }
359 var gotText string
360 var done EventDone
361 for ev := range evs {
362 switch e := ev.(type) {
363 case EventTextDelta:
364 gotText += e.Delta
365 case EventDone:
366 done = e
367 }
368 }
369 if gotText != "hi" {
370 t.Fatalf("text=%q", gotText)
371 }
372 if done.Stop != StopEnd {
373 t.Fatalf("stop=%v", done.Stop)
374 }
375}
376
377func TestClaudeOpus5Catalog(t *testing.T) {
378 m, err := FindModel("anthropic", "claude-opus-5")
379 if err != nil {
380 t.Fatal(err)
381 }
382 if m.DisplayName != "Claude Opus 5" || m.ContextWindow != 1000000 || m.MaxOutput != 128000 || !m.Reasoning || !m.AdaptiveThinking || m.Speculative {
383 t.Fatalf("unexpected Opus 5 model: %+v", m)
384 }
385 if m.PriceInput != 5 || m.PriceOutput != 25 || m.PriceCacheRead != 0.5 || m.PriceCacheWrite != 6.25 {
386 t.Fatalf("unexpected Opus 5 pricing: %+v", m)
387 }
388}
389
390func TestClaudeSonnet5Catalog(t *testing.T) {
391 m, err := FindModel("anthropic", "claude-sonnet-5")
392 if err != nil {
393 t.Fatal(err)
394 }
395 if m.DisplayName != "Claude Sonnet 5" || m.ContextWindow != 1000000 || m.MaxOutput != 128000 || !m.Reasoning || !m.AdaptiveThinking {
396 t.Fatalf("unexpected Sonnet 5 model: %+v", m)
397 }
398 if m.PriceInput != 2 || m.PriceOutput != 10 || m.PriceCacheRead != 0.2 || m.PriceCacheWrite != 2.5 {
399 t.Fatalf("unexpected Sonnet 5 pricing: %+v", m)
400 }
401}
402
403func TestClaudeFable5Catalog(t *testing.T) {
404 m, err := FindModel("anthropic", "claude-fable-5")
405 if err != nil {
406 t.Fatal(err)
407 }
408 if m.DisplayName != "Claude Fable 5" || m.ContextWindow != 1000000 || m.MaxOutput != 128000 || !m.Reasoning || !m.AdaptiveThinking || m.Speculative {
409 t.Fatalf("unexpected Fable 5 model: %+v", m)
410 }
411 if m.PriceInput != 10 || m.PriceOutput != 50 || m.PriceCacheRead != 1 || m.PriceCacheWrite != 12.5 {
412 t.Fatalf("unexpected Fable 5 pricing: %+v", m)
413 }
414}
415
416func TestClaudeFable51Catalog(t *testing.T) {
417 m, err := FindModel("anthropic", "claude-fable-5-1")
418 if err != nil {
419 t.Fatal(err)
420 }
421 if m.DisplayName != "Claude Fable 5.1" || m.ContextWindow != 1000000 || m.MaxOutput != 128000 || !m.Reasoning || !m.AdaptiveThinking || m.Speculative {
422 t.Fatalf("unexpected Fable 5.1 model: %+v", m)
423 }
424 if m.PriceInput != 10 || m.PriceOutput != 50 || m.PriceCacheRead != 0.25 || m.PriceCacheWrite != 12.5 {
425 t.Fatalf("unexpected Fable 5.1 pricing: %+v", m)
426 }
427
428 wire, err := NewAnthropic("x", "").(*anthropicClient).buildRequest(Request{
429 Model: m.ID,
430 Reasoning: "max",
431 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
432 })
433 if err != nil {
434 t.Fatal(err)
435 }
436 if wire.Thinking == nil || wire.Thinking.Type != "adaptive" || wire.Thinking.BudgetTokens != 0 {
437 t.Fatalf("unexpected Fable 5.1 thinking config: %+v", wire.Thinking)
438 }
439 if wire.OutputConfig == nil || wire.OutputConfig.Effort != "max" {
440 t.Fatalf("unexpected Fable 5.1 output config: %+v", wire.OutputConfig)
441 }
442}
443
444func TestGondolaSeedCatalog(t *testing.T) {
445 kimi, err := FindModel("gondola", "kimi-k3")
446 if err != nil {
447 t.Fatal(err)
448 }
449 if kimi.ContextWindow != 1000000 || kimi.MaxOutput != 131072 || !kimi.Reasoning || kimi.AdaptiveThinking {
450 t.Fatalf("unexpected Kimi K3 model: %+v", kimi)
451 }
452 if kimi.PriceInput != 1.1475 || kimi.PriceOutput != 5.7375 || kimi.PriceCacheRead != 0.11475 {
453 t.Fatalf("unexpected Kimi K3 pricing: %+v", kimi)
454 }
455
456 opus, err := FindModel("gondola", "claude-opus-5")
457 if err != nil {
458 t.Fatal(err)
459 }
460 if !opus.AdaptiveThinking || opus.PriceInput != 1.836 || opus.PriceOutput != 9.18 || opus.PriceCacheRead != 0.1836 || opus.PriceCacheWrite != 2.295 {
461 t.Fatalf("unexpected Claude Opus 5 model: %+v", opus)
462 }
463}
464
465func TestOpenAICompatAnthropicReasoningEffort(t *testing.T) {
466 c := NewOpenRouter("token", "").(*openaiClient)
467 wire, err := c.buildRequest(Request{
468 Model: "anthropic/claude-opus-4.8",
469 Reasoning: "maximum",
470 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
471 })
472 if err != nil {
473 t.Fatal(err)
474 }
475 if wire.ReasoningEffort != "xhigh" {
476 t.Fatalf("want xhigh for adaptive Anthropic model over OpenAI-compatible wire, got %q", wire.ReasoningEffort)
477 }
478
479 wire, err = c.buildRequest(Request{
480 Model: "anthropic/claude-sonnet-5",
481 Reasoning: "maximum",
482 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
483 })
484 if err != nil {
485 t.Fatal(err)
486 }
487 if wire.ReasoningEffort != "xhigh" {
488 t.Fatalf("want xhigh for Claude Sonnet 5 over OpenAI-compatible wire, got %q", wire.ReasoningEffort)
489 }
490
491 wire, err = c.buildRequest(Request{
492 Model: "gpt-5.1",
493 Reasoning: "maximum",
494 Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
495 })
496 if err != nil {
497 t.Fatal(err)
498 }
499 if wire.ReasoningEffort != "high" {
500 t.Fatalf("want generic OpenAI-compatible maximum clamped to high, got %q", wire.ReasoningEffort)
501 }
502}
503
504func TestOpenAIBuildRequestSkipsReasoningOnlyAssistantMessages(t *testing.T) {
505 c := NewKimi("token", "").(*openaiClient)
506 wire, err := c.buildRequest(Request{
507 Model: "kimi-for-coding",
508 Messages: []Message{
509 {Role: RoleUser, Content: []Content{TextBlock{Text: "first"}}},
510 {Role: RoleAssistant, Content: []Content{ReasoningBlock{Summary: "thinking only"}}},
511 {Role: RoleUser, Content: []Content{TextBlock{Text: "second"}}},
512 },
513 })
514 if err != nil {
515 t.Fatal(err)
516 }
517 for i, msg := range wire.Messages {
518 if msg.Role == "assistant" && msg.Content == nil && len(msg.ToolCalls) == 0 {
519 t.Fatalf("message %d is empty assistant: %+v", i, msg)
520 }
521 }
522 if got := len(wire.Messages); got != 2 {
523 t.Fatalf("messages=%d want 2 after skipping reasoning-only assistant", got)
524 }
525}
526
527func TestOpenAIStreamHappyPath(t *testing.T) {
528 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
529 w.Header().Set("content-type", "text/event-stream")
530 w.WriteHeader(http.StatusOK)
531 fl, _ := w.(http.Flusher)
532 write := func(s string) {
533 _, _ = w.Write([]byte(s))
534 if fl != nil {
535 fl.Flush()
536 }
537 }
538 write("data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"hel\"},\"finish_reason\":null}]}\n\n")
539 write("data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"lo\"},\"finish_reason\":\"stop\"}]}\n\n")
540 write("data: {\"choices\":[],\"usage\":{\"prompt_tokens\":8,\"completion_tokens\":2}}\n\n")
541 write("data: [DONE]\n\n")
542 }))
543 defer srv.Close()
544
545 c := NewOpenAI("x", srv.URL)
546 evs, err := c.Stream(context.Background(), Request{Model: "gpt-5"})
547 if err != nil {
548 t.Fatal(err)
549 }
550 var gotText string
551 var done EventDone
552 for ev := range evs {
553 switch e := ev.(type) {
554 case EventTextDelta:
555 gotText += e.Delta
556 case EventDone:
557 done = e
558 }
559 }
560 if gotText != "hello" {
561 t.Fatalf("text=%q", gotText)
562 }
563 if done.Stop != StopEnd {
564 t.Fatalf("stop=%v", done.Stop)
565 }
566}
567
568func TestDiscoverOpenRouter(t *testing.T) {
569 const body = `{"data":[
570 {"id":"x/full","pricing":{"prompt":"0.000003","completion":"0.000015"},
571 "context_length":200000,"top_provider":{"max_completion_tokens":64000},
572 "supported_parameters":["reasoning"]},
573 {"id":"x/fallback","top_provider":{"context_length":128000}},
574 {"id":""}]}`
575 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
576 _, _ = w.Write([]byte(body))
577 }))
578 defer srv.Close()
579
580 models, err := DiscoverOpenRouter(context.Background(), srv.URL)
581 if err != nil {
582 t.Fatal(err)
583 }
584 if len(models) != 2 { // empty id dropped
585 t.Fatalf("want 2 models, got %d", len(models))
586 }
587 // per-token USD -> per-1M; reasoning from supported_parameters.
588 if m := models[0]; m.Provider != "openrouter" || m.ContextWindow != 200000 ||
589 m.MaxOutput != 64000 || !m.Reasoning || m.PriceInput != 3 || m.PriceOutput != 15 {
590 t.Errorf("model[0]: %+v", m)
591 }
592 // context falls back to top_provider; no reasoning.
593 if m := models[1]; m.ContextWindow != 128000 || m.MaxOutput != 0 || m.Reasoning {
594 t.Errorf("model[1]: %+v", m)
595 }
596}
597
598func TestGPT56CatalogEntries(t *testing.T) {
599 cases := []struct {
600 provider string
601 id string
602 context int
603 priceInput float64
604 priceOut float64
605 cacheRead float64
606 cacheWrite float64
607 }{
608 {"openai", "gpt-5.6-luna", 272000, 1, 6, 0.1, 1.25},
609 {"openai", "gpt-5.6-sol", 272000, 5, 30, 0.5, 6.25},
610 {"openai", "gpt-5.6-terra", 272000, 2.5, 15, 0.25, 3.125},
611 {"openai-codex", "gpt-5.6-luna", 272000, 1, 6, 0.1, 1.25},
612 {"openai-codex", "gpt-5.6-sol", 272000, 5, 30, 0.5, 6.25},
613 {"openai-codex", "gpt-5.6-terra", 272000, 2.5, 15, 0.25, 3.125},
614 {"azure-openai-responses", "gpt-5.6-luna", 1050000, 1, 6, 0.1, 1.25},
615 {"azure-openai-responses", "gpt-5.6-sol", 1050000, 5, 30, 0.5, 6.25},
616 {"azure-openai-responses", "gpt-5.6-terra", 1050000, 2.5, 15, 0.25, 3.125},
617 {"vercel-ai-gateway", "openai/gpt-5.6-luna", 1050000, 1, 6, 0.1, 1.25},
618 {"vercel-ai-gateway", "openai/gpt-5.6-sol", 1050000, 5, 30, 0.5, 6.25},
619 {"vercel-ai-gateway", "openai/gpt-5.6-terra", 1050000, 2.5, 15, 0.25, 3.125},
620 }
621 for _, tc := range cases {
622 m, err := FindModel(tc.provider, tc.id)
623 if err != nil {
624 t.Fatal(err)
625 }
626 if m.ContextWindow != tc.context || m.MaxOutput != 128000 || !m.Reasoning {
627 t.Fatalf("%s/%s limits: %+v", tc.provider, tc.id, m)
628 }
629 if m.PriceInput != tc.priceInput || m.PriceOutput != tc.priceOut || m.PriceCacheRead != tc.cacheRead || m.PriceCacheWrite != tc.cacheWrite {
630 t.Fatalf("%s/%s prices: %+v", tc.provider, tc.id, m)
631 }
632 }
633}
634
635// TestOpenCodeGoCatalog pins entries added from the vendor's current Go
636// model lineup (https://opencode.ai/docs/go).
637func TestOpenCodeGoCatalog(t *testing.T) {
638 cases := []struct {
639 id string
640 context int
641 maxOut int
642 priceIn float64
643 priceOut float64
644 cacheRead float64
645 }{
646 {"gpt-5.6-luna", 1050000, 128000, 0.2, 1.2, 0.02},
647 {"grok-4.5", 500000, 500000, 2, 6, 0.3},
648 {"kimi-k3", 1048576, 131072, 3, 15, 0.3},
649 }
650 for _, tc := range cases {
651 m, err := FindModel("opencode-go", tc.id)
652 if err != nil {
653 t.Fatal(err)
654 }
655 if m.ContextWindow != tc.context || m.MaxOutput != tc.maxOut || !m.Reasoning {
656 t.Fatalf("opencode-go/%s limits: %+v", tc.id, m)
657 }
658 if m.PriceInput != tc.priceIn || m.PriceOutput != tc.priceOut || m.PriceCacheRead != tc.cacheRead {
659 t.Fatalf("opencode-go/%s prices: %+v", tc.id, m)
660 }
661 if m.BaseURL != "https://opencode.ai/zen/go/v1" {
662 t.Fatalf("opencode-go/%s baseURL: %s", tc.id, m.BaseURL)
663 }
664 }
665
666 luna, err := FindModel("opencode-go", "gpt-5.6-luna")
667 if err != nil {
668 t.Fatal(err)
669 }
670 if luna.API != APIResponses || luna.PriceCacheWrite != 0.25 {
671 t.Fatalf("opencode-go/gpt-5.6-luna API or cache-write price: %+v", luna)
672 }
673 if luna.PriceTierInputTokens != 272000 || luna.PriceInputAbove != 0.4 || luna.PriceOutputAbove != 1.8 || luna.PriceCacheReadAbove != 0.04 || luna.PriceCacheWriteAbove != 0.5 {
674 t.Fatalf("opencode-go/gpt-5.6-luna long-context prices: %+v", luna)
675 }
676}
677
678// TestOpenAIOmitsZeroMaxTokens guards against sending max_tokens: 0 for
679// discovered models that don't advertise an output cap (MaxOutput == 0).
680func TestOpenAIOmitsZeroMaxTokens(t *testing.T) {
681 SetLiveModels([]Model{
682 {Provider: "openrouter", ID: "vendor/no-cap", DisplayName: "No Cap"},
683 {Provider: "openrouter", ID: "vendor/reason-no-cap", DisplayName: "Reason No Cap", Reasoning: true},
684 {Provider: "openrouter", ID: "vendor/capped", DisplayName: "Capped", MaxOutput: 4096},
685 })
686 defer SetLiveModels(nil)
687
688 c := NewOpenAI("x", "").(*openaiClient)
689
690 got, err := c.buildRequest(Request{Model: "vendor/no-cap"})
691 if err != nil {
692 t.Fatal(err)
693 }
694 if got.MaxTokens != nil {
695 t.Errorf("expected max_tokens omitted, got %d", *got.MaxTokens)
696 }
697
698 got, err = c.buildRequest(Request{Model: "vendor/reason-no-cap"})
699 if err != nil {
700 t.Fatal(err)
701 }
702 if got.MaxCompletionTok != nil {
703 t.Errorf("expected max_completion_tokens omitted, got %d", *got.MaxCompletionTok)
704 }
705
706 got, err = c.buildRequest(Request{Model: "vendor/capped"})
707 if err != nil {
708 t.Fatal(err)
709 }
710 if got.MaxTokens == nil || *got.MaxTokens != 4096 {
711 t.Errorf("expected max_tokens 4096, got %v", got.MaxTokens)
712 }
713}