codeep 3.2.2 → 3.3.0

package/dist/config/providers.js

@@ -19,7 +19,7 @@ export const PROVIDERS = {
         },
         models: [
             { id: 'glm-5.3', name: 'GLM-5.3', description: 'Latest flagship for project-scale engineering (1M context)' },
-            { id: 'glm-5.3-flash', name: 'GLM-5.3 Flash', description: 'Same 1M context at a twentieth of the price' },
+            { id: 'glm-5.3-flash', name: 'GLM-5.3 Flash', description: 'Same 1M context at roughly a ninth of the price' },
             { id: 'glm-5.2', name: 'GLM-5.2', description: 'Previous flagship for project-scale engineering (1M context)' },
             { id: 'glm-5-turbo', name: 'GLM-5 Turbo', description: 'Fast GLM-5 variant, available to all users' },
         ],
@@ -53,7 +53,7 @@ export const PROVIDERS = {
             // (`z.ai-cn*`) bills separately and is not covered by that page, so it is
             // left alone until its own listing is checked.
             { id: 'glm-5.3', name: 'GLM-5.3', description: 'Latest flagship for project-scale engineering (1M context)' },
-            { id: 'glm-5.3-flash', name: 'GLM-5.3 Flash', description: 'Same 1M context at a twentieth of the price' },
+            { id: 'glm-5.3-flash', name: 'GLM-5.3 Flash', description: 'Same 1M context at roughly a ninth of the price' },
             { id: 'glm-5.2', name: 'GLM-5.2', description: 'Previous flagship for project-scale engineering (1M context)' },
             { id: 'glm-5-turbo', name: 'GLM-5 Turbo', description: 'Fast GLM-5 variant' },
         ],
@@ -80,11 +80,14 @@ export const PROVIDERS = {
                 supportsNativeTools: true,
             },
         },
+        // GLM-5.3 is on the China gateway and the China Coding Plan (bigmodel.cn).
+        // Flash is listed only on China pay-per-use, so the plan does not offer it.
         models: [
-            { id: 'glm-5.2', name: 'GLM-5.2', description: 'Latest flagship for project-scale engineering (1M context)' },
+            { id: 'glm-5.3', name: 'GLM-5.3', description: 'Latest flagship for project-scale engineering (1M context)' },
+            { id: 'glm-5.2', name: 'GLM-5.2', description: 'Previous flagship for project-scale engineering (1M context)' },
             { id: 'glm-5-turbo', name: 'GLM-5 Turbo', description: 'Fast GLM-5 variant, available to all users' },
         ],
-        defaultModel: 'glm-5.2',
+        defaultModel: 'glm-5.3',
         defaultProtocol: 'openai',
         maxOutputTokens: 131_072,
         envKey: 'ZAI_CN_API_KEY',
@@ -109,10 +112,12 @@ export const PROVIDERS = {
             },
         },
         models: [
-            { id: 'glm-5.2', name: 'GLM-5.2', description: 'Latest flagship for project-scale engineering (1M context)' },
+            { id: 'glm-5.3', name: 'GLM-5.3', description: 'Latest flagship for project-scale engineering (1M context)' },
+            { id: 'glm-5.3-flash', name: 'GLM-5.3 Flash', description: 'Same 1M context at a tenth of the price' },
+            { id: 'glm-5.2', name: 'GLM-5.2', description: 'Previous flagship for project-scale engineering (1M context)' },
             { id: 'glm-5-turbo', name: 'GLM-5 Turbo', description: 'Fast GLM-5 variant' },
         ],
-        defaultModel: 'glm-5.2',
+        defaultModel: 'glm-5.3',
         defaultProtocol: 'openai',
         maxOutputTokens: 131_072,
         envKey: 'ZAI_CN_API_KEY',
@@ -207,11 +212,19 @@ export const PROVIDERS = {
                 supportsNativeTools: true,
             },
         },
+        // One model. DeepSeek's own pricing page names `deepseek-flash` (V4.1 Flash)
+        // as the model to use and says it outperforms V4 Pro. `deepseek-v4-flash`
+        // is already served by it, and from 2026-09-14 12:00 Beijing time every
+        // `deepseek-v4-pro` request is routed to it and billed at its price until
+        // a V4.1 Pro exists — so a Pro entry would be a label on a different model.
+        // Both old ids migrate via RETIRED_MODEL_REPLACEMENTS.
+        //
+        // Note the id differs on OpenRouter, which lists the same model as
+        // `deepseek/deepseek-v4.1-flash`.
         models: [
-            { id: 'deepseek-v4-pro', name: 'DeepSeek V4 Pro', description: 'Most capable DeepSeek model (1M context, thinking mode)' },
-            { id: 'deepseek-v4-flash', name: 'DeepSeek V4 Flash', description: 'Fast and affordable DeepSeek model (1M context)' },
+            { id: 'deepseek-flash', name: 'DeepSeek V4.1 Flash', description: 'Current DeepSeek model — thinking on by default, vision, 1M context, 384K output' },
         ],
-        defaultModel: 'deepseek-v4-pro',
+        defaultModel: 'deepseek-flash',
         defaultProtocol: 'openai',
         maxOutputTokens: 384_000, // DeepSeek V4 max output
         envKey: 'DEEPSEEK_API_KEY',
@@ -344,12 +357,17 @@ export const PROVIDERS = {
         protocols: {
             openai: { baseUrl: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1', authHeader: 'Bearer', supportsNativeTools: true },
         },
+        // 3.8 Max and Flash are GA on Model Studio international and cost less
+        // than the 3.7 Max / 3.6 Flash they sit above. China (`qwen-cn-api`) is a
+        // separate listing and is not changed on this evidence.
         models: [
-            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Latest flagship for complex coding and reasoning' },
+            { id: 'qwen3.8-max', name: 'Qwen3.8-Max', description: 'Latest flagship for complex coding and reasoning (1M context)' },
+            { id: 'qwen3.8-flash', name: 'Qwen3.8-Flash', description: 'Low-latency, low-cost multimodal model (1M context)' },
+            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Previous flagship for complex coding and reasoning' },
             { id: 'qwen3.7-plus', name: 'Qwen3.7-Plus', description: 'Balanced quality, speed, and price (1M context)' },
-            { id: 'qwen3.6-flash', name: 'Qwen3.6-Flash', description: 'Low-latency, low-cost multimodal model' },
+            { id: 'qwen3.6-flash', name: 'Qwen3.6-Flash', description: 'Previous low-latency, low-cost multimodal model' },
         ],
-        defaultModel: 'qwen3.7-max',
+        defaultModel: 'qwen3.8-max',
         defaultProtocol: 'openai',
         maxOutputTokens: 65_536,
         noStreamWithTools: true,
@@ -535,12 +553,17 @@ export const PROVIDERS = {
             { id: 'anthropic/claude-fable-5', name: 'Claude Fable 5', description: 'Anthropic — superseded by 5.1' },
             { id: 'anthropic/claude-opus-5', name: 'Claude Opus 5', description: 'Anthropic — flagship Opus tier' },
             { id: 'anthropic/claude-sonnet-5', name: 'Claude Sonnet 5', description: 'Anthropic — balanced' },
+            { id: 'openai/gpt-6-astra', name: 'GPT-6 Astra', description: 'OpenAI — frontier' },
             { id: 'openai/gpt-5.6-sol', name: 'GPT-5.6 Sol', description: 'OpenAI — flagship' },
             { id: 'openai/gpt-5.6-luna', name: 'GPT-5.6 Luna', description: 'OpenAI — fast/efficient' },
-            { id: 'google/gemini-3.7-flash', name: 'Gemini 3.7 Flash', description: 'Google — latest production Flash' },
-            { id: 'deepseek/deepseek-v4-pro', name: 'DeepSeek V4 Pro', description: 'DeepSeek — flagship agentic model' },
+            { id: 'google/gemini-3.8-flash', name: 'Gemini 3.8 Flash', description: 'Google — latest production Flash' },
+            { id: 'google/gemini-3.7-flash', name: 'Gemini 3.7 Flash', description: 'Google — previous production Flash' },
+            { id: 'deepseek/deepseek-v4.1-flash', name: 'DeepSeek V4.1 Flash', description: 'DeepSeek — current model' },
             { id: 'moonshotai/kimi-k3', name: 'Kimi K3', description: 'Moonshot — long-horizon coding' },
-            { id: 'qwen/qwen3.8-max', name: 'Qwen 3.8 Max', description: 'Alibaba — latest flagship' },
+            // Dated on purpose: OpenRouter lists only the snapshot. The undated
+            // `qwen/qwen3.8-max` this used to carry does not exist there, so picking
+            // it before the live catalogue loaded was an error.
+            { id: 'qwen/qwen3.8-max-0902', name: 'Qwen 3.8 Max', description: 'Alibaba — latest flagship' },
             { id: 'x-ai/grok-4.6', name: 'Grok 4.6', description: 'xAI — flagship reasoning' },
         ],
         defaultModel: 'openrouter/auto',
@@ -607,6 +630,13 @@ const RETIRED_MODEL_REPLACEMENTS = {
     google: {
         'gemini-3.1-flash-lite': 'gemini-3.5-flash-lite',
     },
+    // DeepSeek named V4.1 Flash the model to use: the V4 Flash ids are retired
+    // and served by it, and V4 Pro is routed to it from 2026-09-14.
+    deepseek: {
+        'deepseek-v4-flash': 'deepseek-flash',
+        'deepseek-v4-flash-vision-exp': 'deepseek-flash',
+        'deepseek-v4-pro': 'deepseek-flash',
+    },
     grok: {
         'grok-code-fast-1': 'grok-build-0.1',
         'grok-4-fast-reasoning': 'grok-4.3',
@@ -848,7 +878,10 @@ export function modelSupportsReasoningEffort(providerId, model) {
             // Gemini 3.x thinking_level via the OpenAI-compat reasoning_effort mapping.
             return id.startsWith('gemini-3');
         case 'deepseek':
-            return id.startsWith('deepseek-v4');
+            // `deepseek-flash` does not start with `deepseek-v4` — a prefix check
+            // alone would ship the current model with /thinking hidden, the same
+            // trap GPT-6 nearly fell into behind `gpt-5`.
+            return id.startsWith('deepseek-v4') || idMatches(id, 'deepseek-flash');
         case 'z.ai':
         case 'z.ai-api':
         case 'z.ai-cn':
@@ -908,7 +941,12 @@ export function reasoningParamsFor(providerId, model, tier) {
             // 'minimal' is deliberately not emitted: 3.7 Flash rejects it outright.
             return { reasoning_effort: tier === 'max' ? 'high' : tier };
         case 'deepseek':
-            // Graded thinking depth: high (default) or max. Lower tiers collapse to high.
+            // V4.1 Flash distinguishes low / high / max (DeepSeek's mapping table:
+            // minimal+low→low, medium+high+xhigh→high, max+ultra→max). The retired V4
+            // ids graded only high|max. Always an effort, never a disabled block.
+            if (idMatches(canonicalModelId(model), 'deepseek-flash')) {
+                return { reasoning_effort: tier === 'low' ? 'low' : tier === 'max' ? 'max' : 'high' };
+            }
             return { reasoning_effort: tier === 'max' ? 'max' : 'high' };
         case 'z.ai':
         case 'z.ai-api':
@@ -959,7 +997,9 @@ export function availableReasoningTiers(providerId, model) {
             // coding, so collapsing it hid the setting most users want.
             return ['auto', 'low', 'medium', 'high'];
         case 'deepseek':
-            return ['auto', 'high', 'max'];
+            return idMatches(canonicalModelId(model), 'deepseek-flash')
+                ? ['auto', 'low', 'high', 'max']
+                : ['auto', 'high', 'max'];
         case 'z.ai':
         case 'z.ai-api':
         case 'z.ai-cn':

package/dist/utils/tokenTracker.js

@@ -30,6 +30,7 @@ const MODEL_CONTEXT_WINDOWS = {
     'claude-sonnet-5': 1_000_000,
     'claude-haiku-4-5-20251001': 200_000,
     // DeepSeek
+    'deepseek-flash': 1_000_000,
     'deepseek-v4-pro': 1_000_000,
     'deepseek-v4-flash': 1_000_000,
     // Google
@@ -43,7 +44,7 @@ const MODEL_CONTEXT_WINDOWS = {
     // MiniMax
     'MiniMax-M3': 1_000_000,
     // Kimi (Moonshot) — 1M on K3, 256K across K2.x
-    'kimi-k3': 1_000_000,
+    'kimi-k3': 1_048_576,
     'kimi-k2.7-code': 262_144,
     'kimi-k2.7-code-highspeed': 262_144,
     'kimi-k2.6': 262_144,
@@ -59,6 +60,8 @@ const MODEL_CONTEXT_WINDOWS = {
     'grok-code-fast-1': 256_000,
     'grok-4-fast-reasoning': 2_000_000,
     // Qwen (Alibaba) — current hosted generation
+    'qwen3.8-max': 1_000_000,
+    'qwen3.8-flash': 1_000_000,
     'qwen3.7-max': 1_000_000,
     'qwen3.8-max-preview': 1_000_000,
     'qwen3.7-plus': 1_000_000,
@@ -93,10 +96,14 @@ const MODEL_PRICING = {
     'glm-5.1': { inputPer1M: 1.40, outputPer1M: 4.40 },
     'glm-5-turbo': { inputPer1M: 1.20, outputPer1M: 4.00 },
     // OpenAI
-    // Twice the price of 5.6 Sol. Cached reads are $1.00/M — 0.1x input, which is
+    // 2.5x 5.6 Sol at its current promotional rate. Cached reads are $1.00/M — 0.1x input, which is
     // what DEFAULT_CACHE_READ_RATE already applies, so no model row is needed.
     'gpt-6-astra': { inputPer1M: 10.00, outputPer1M: 50.00 },
-    'gpt-5.6-sol': { inputPer1M: 5.00, outputPer1M: 30.00 },
+    // Promotional rate "available at least through 2026-11-21" per OpenAI's
+    // pricing page. Carried because it is what users are billed now, and no end
+    // date is written ahead of time — the Sonnet 5 lesson. Long-context requests
+    // bill 8/30; like every row here this holds the standard short-context tier.
+    'gpt-5.6-sol': { inputPer1M: 4.00, outputPer1M: 20.00 },
     'gpt-5.6-terra': { inputPer1M: 2.00, outputPer1M: 12.00 },
     'gpt-5.6-luna': { inputPer1M: 0.20, outputPer1M: 1.20 },
     'gpt-5.5': { inputPer1M: 5.00, outputPer1M: 30.00 },
@@ -116,8 +123,19 @@ const MODEL_PRICING = {
     // honest direction for a cost estimate, and rule 5 of the catalogue policy
     // allows a clearly-labelled conservative approximation but never an invented
     // number. Cache-miss input; cache hits are ~1/50th and not modelled here.
-    'deepseek-v4-pro': { inputPer1M: 0.435, outputPer1M: 0.87 },
-    'deepseek-v4-flash': { inputPer1M: 0.14, outputPer1M: 0.28 },
+    // DeepSeek bills peak / off-peak (off-peak is half), and this table holds one
+    // rate, so it carries PEAK — an over-estimate by design. The previous rows
+    // (0.435/0.87 and 0.14/0.28) were two to four and a half times below today's
+    // peak and so under-reported, which is the one direction this table must not
+    // err in. Cache hits cost 2% of a miss, but DeepSeek reports them as
+    // `prompt_cache_hit_tokens`, which this tracker does not read — so every input
+    // token bills at the miss rate: an over-estimate again, not a gap.
+    'deepseek-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
+    // Retired V4 Flash is served by V4.1 Flash and billed at its price.
+    'deepseek-v4-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
+    // Pro's own peak rate, right for every run until 2026-09-14; after that the id
+    // is routed to Flash, and configs holding it have been migrated away from it.
+    'deepseek-v4-pro': { inputPer1M: 1.32, outputPer1M: 3.96 },
     // Google
     // Gemini 3.6/3.7/3.8 Flash carry PROMOTIONAL rates that run through 2026-12-31
     // and are scheduled to step up to 1.50/7.50 on 2027-01-01 — revisit all three
@@ -132,7 +150,10 @@ const MODEL_PRICING = {
     'gemini-3.5-flash-lite': { inputPer1M: 0.30, outputPer1M: 2.50 },
     'gemini-3-flash-preview': { inputPer1M: 0.50, outputPer1M: 3.00 },
     // MiniMax
-    'MiniMax-M3': { inputPer1M: 0.60, outputPer1M: 2.40 },
+    // "Permanent 50% off" list price for prompts up to 512K tokens; above that,
+    // 0.60/2.40. The old 0.60/2.40 row was the long-context tier and doubled the
+    // estimate for virtually every real request.
+    'MiniMax-M3': { inputPer1M: 0.30, outputPer1M: 1.20 },
     // Kimi (Moonshot) — pay-per-use cache-miss rates; `kimi-for-coding` is the
     // subscription alias (flat-fee in reality, priced notionally like K2.7 Code).
     'kimi-k3': { inputPer1M: 3.00, outputPer1M: 15.00 },
@@ -160,6 +181,8 @@ const MODEL_PRICING = {
     // `qwen3.8-max-preview` is Token-Plan-only (credit-metered, promotional
     // preview rate); Alibaba publishes no pay-per-use per-token price for it.
     // Leave it unpriced rather than borrowing the GA qwen3.8-max rate.
+    'qwen3.8-max': { inputPer1M: 2.00, outputPer1M: 6.00 },
+    'qwen3.8-flash': { inputPer1M: 0.15, outputPer1M: 0.47 },
     'qwen3.7-max': { inputPer1M: 2.50, outputPer1M: 7.50 },
     'qwen3.7-plus': { inputPer1M: 0.40, outputPer1M: 1.60 },
     'qwen3.6-plus': { inputPer1M: 0.40, outputPer1M: 2.40 },

package/dist/version.d.ts

@@ -1 +1 @@
-export declare const VERSION = "3.2.2";
+export declare const VERSION = "3.3.0";

package/dist/version.js

@@ -1,4 +1,4 @@
 // AUTO-GENERATED by scripts/gen-version.js — do not edit by hand.
 // Baked from package.json at build time so the bun-compiled binary reports
 // the right version (it has no package.json on disk to read at runtime).
-export const VERSION = '3.2.2';
+export const VERSION = '3.3.0';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeep",
-  "version": "3.2.2",
+  "version": "3.3.0",
   "description": "AI-powered coding assistant built for the terminal. Multiple LLM providers, project-aware context, and a seamless development workflow.",
   "type": "module",
   "main": "dist/index.js",