zerogpu · amaan-ai20 · Sep 14, 2026
diff --git a/README.md b/README.md
@@ -21,7 +21,6 @@ The official command-line interface for [ZeroGPU](https://zerogpu.ai) — run fa
     - [`chat`](#chat)
     - [`chat_thinking`](#chat_thinking)
     - [`summarize`](#summarize)
-    - [`generate_followups`](#generate_followups)
   - [Classification](#classification)
     - [`classify_iab`](#classify_iab)
     - [`classify_iab_enriched`](#classify_iab_enriched)
@@ -135,11 +134,11 @@ zerogpu chat "Why does my API keep getting rate-limited?" -m gpt-oss-120b -r
 # Multilingual reasoning
 zerogpu chat "Explique la mise en cache en une phrase." -m qwen3-30b-a3b-fp8
 
-# A 1M-token context, for whole repos and very long documents
+# The platform's most capable model, and its priciest
 zerogpu chat "$(cat ARCHITECTURE.md)" -m glm-5.2
 
 # Coding and agentic work, at a fraction of the flagship price
-zerogpu chat "Port this helper to async/await." -m deepseek-v4-flash
+zerogpu chat "Port this helper to async/await." -m deepseek-v4-flash-0731
 ```
 
 | Option | Description |
@@ -152,12 +151,12 @@ zerogpu chat "Port this helper to async/await." -m deepseek-v4-flash
 |---|---|
 | `LFM2.5-1.2B-Instruct` | Default. Fast edge chat. |
 | `LFM2.5-1.2B-Thinking` | Compact reasoning model. |
-| `gpt-oss-120b` | 117B MoE, 131K context, reasoning + function calling. |
-| `qwen3-30b-a3b-fp8` | 30.5B MoE, 100+ languages, reasoning + function calling. |
-| `glm-5.2` | 753B MoE, 1M context, reasoning + function calling. The platform's most capable model, and its priciest. |
-| `deepseek-v4-flash` | 284B MoE (13B active), 1M context, coding and agentic workflows. |
+| `gpt-oss-120b` | 120B MoE, 131K context, reasoning + function calling. |
+| `qwen3-30b-a3b-fp8` | 30B MoE, 100+ languages, reasoning + function calling. |
+| `glm-5.2` | 753B MoE, 262K context, reasoning + function calling. The platform's most capable model, and its priciest. |
+| `deepseek-v4-flash-0731` | 284B MoE (13B active), 1M context, coding and agentic workflows. |
 
-`qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash` are served by the Chat Completions API rather than the Responses API; the CLI routes them automatically.
+`qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash-0731` are served by the Chat Completions API rather than the Responses API; the CLI routes them automatically.
 
 #### `chat_thinking`
 
@@ -175,14 +174,6 @@ Summarize text with the **llama-3.1-8b-instruct-fast** model.
 zerogpu summarize "Long article text goes here..."
 ```
 
-#### `generate_followups`
-
-Generate contextual follow-up questions using the ZeroGPU follow-up edge model.
-
-```bash
-zerogpu generate_followups "We just shipped a new pricing page focused on enterprise plans."
-```
-
 ---
 
 ### Classification

diff --git a/docs/ADDING_COMMANDS.md b/docs/ADDING_COMMANDS.md
@@ -7,7 +7,7 @@ This guide explains how to add a new CLI command to the ZeroGPU CLI.
 - `src/commands/` — one file per command, each exporting a `register<Name>Command(program)` function.
 - `src/cli.ts` — wires every command into the root program.
 - `src/lib/responses.ts` — shared `RESPONSES_ENDPOINT`, `ResponsesApiResponse`, and the `extractOutputText` / `extractReasoningText` helpers for `/v1/responses` calls.
-- `src/lib/chatCompletions.ts` — the same for `/v1/chat/completions`, used by models the platform serves only there (currently `qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash`), plus `toResponsesUsage` to normalize token counts for savings tracking.
+- `src/lib/chatCompletions.ts` — the same for `/v1/chat/completions`, used by models the platform serves only there (currently `qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash-0731`), plus `toResponsesUsage` to normalize token counts for savings tracking.
 - `src/lib/auth.ts` — `getApiKey()` for authenticated requests.
 
 ## Steps

diff --git a/docs/DOCUMENTATION.md b/docs/DOCUMENTATION.md
@@ -5,15 +5,14 @@
 `zerogpu-cli` is the official command-line interface for [ZeroGPU](https://zerogpu.ai), a distributed / edge inference platform for small language models (SLMs) and nano language models. The CLI is a thin, OpenAI-compatible client around the ZeroGPU **Responses API** (`https://api.zerogpu.ai/v1/responses`) — and, for models served only there, the **Chat Completions API** (`https://api.zerogpu.ai/v1/chat/completions`) — that lets you call a curated set of edge-optimized models directly from your terminal for common NLP workloads:
 
 - Conversational chat (`LFM2.5-1.2B-Instruct`, `LFM2.5-1.2B-Thinking`)
-- Reasoning and tool-use chat (`gpt-oss-120b`, `qwen3-30b-a3b-fp8`, `glm-5.2`, `deepseek-v4-flash`)
+- Reasoning and tool-use chat (`gpt-oss-120b`, `qwen3-30b-a3b-fp8`, `glm-5.2`, `deepseek-v4-flash-0731`)
 - IAB content/audience classification (`zlm-v1-iab-classify-edge`, `zlm-v2-iab-classify-edge-enriched`)
 - Domain-level IAB classification (`zlm-v1-iab-domain-classifier`)
 - Zero-shot classification (`deberta-v3-small`)
 - Structured / schema-driven classification and JSON extraction (`gliner2-base-v1`)
 - Named-entity recognition with custom labels (`gliner2-base-v1`)
 - PII extraction & redaction (`gliner-multi-pii-v1`)
 - Text summarization (`llama-3.1-8b-instruct-fast`)
-- Follow-up question generation (`zlm-v1-followup-questions-edge`)
 
 It is written in TypeScript (ESM, Node ≥ 20), built on [`commander`](https://github.com/tj/commander.js), and ships a single executable: `zerogpu`.
 
@@ -102,8 +101,7 @@ The CLI exposes the following commands:
 | [`redact_pii`](#411-redact_pii) | Mask PII in-line in the text |
 | [`extract_json`](#412-extract_json) | Schema-driven structured JSON extraction |
 | [`summarize`](#413-summarize) | Summarize text with `llama-3.1-8b-instruct-fast` |
-| [`generate_followups`](#414-generate_followups) | Generate follow-up questions |
-| [`classify_domain`](#415-classify_domain) | Domain-level IAB classification |
+| [`classify_domain`](#414-classify_domain) | Domain-level IAB classification |
 
 ### Common exit codes
 | Code | Meaning |
@@ -207,12 +205,12 @@ zerogpu chat <text> [-i <instructions>] [-m <model>] [-r]
 |---|---|---|
 | `LFM2.5-1.2B-Instruct` | Responses | Default. Fast edge chat. |
 | `LFM2.5-1.2B-Thinking` | Responses | Compact reasoning model. |
-| `gpt-oss-120b` | Responses | 117B MoE, 131K context, reasoning + function calling. |
-| `qwen3-30b-a3b-fp8` | Chat Completions | 30.5B MoE, 100+ languages, reasoning + function calling. |
-| `glm-5.2` | Chat Completions | 753B MoE, 1,048,576-token context, reasoning + function calling. The most capable model on the platform, and the most expensive by an order of magnitude. |
-| `deepseek-v4-flash` | Chat Completions | 284B MoE (13B active), 1,048,576-token context, coding and agentic workflows. |
+| `gpt-oss-120b` | Responses | 120B MoE, 131K context, reasoning + function calling. |
+| `qwen3-30b-a3b-fp8` | Chat Completions | 30B MoE, 100+ languages, reasoning + function calling. |
+| `glm-5.2` | Chat Completions | 753B MoE, 262,144-token context, reasoning + function calling. The most capable model on the platform, and the most expensive by an order of magnitude. |
+| `deepseek-v4-flash-0731` | Chat Completions | 284B MoE (13B active), 1,048,576-token context, coding and agentic workflows. |
 
-`qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash` have no Responses endpoint, so the CLI posts them to `/v1/chat/completions` instead, mapping `--instructions` to a `system` message and normalizing `prompt_tokens` / `completion_tokens` back to Responses token names for savings tracking. This routing is transparent — the command and its output are identical either way.
+`qwen3-30b-a3b-fp8`, `glm-5.2`, and `deepseek-v4-flash-0731` have no Responses endpoint, so the CLI posts them to `/v1/chat/completions` instead, mapping `--instructions` to a `system` message and normalizing `prompt_tokens` / `completion_tokens` back to Responses token names for savings tracking. This routing is transparent — the command and its output are identical either way.
 
 **Example**
 ```bash
@@ -599,41 +597,7 @@ A single condensed summary string.
 
 ---
 
-### 4.14 `generate_followups`
-
-Generate contextual follow-up questions using `zlm-v1-followup-questions-edge`.
-
-**Synopsis**
-```
-zerogpu generate_followups <text>
-```
-
-**Parameters**
-
-| Name | Type | Required | Description |
-|---|---|---|---|
-| `text` (positional) | string | yes | Conversation turn / passage to generate follow-ups for. |
-
-**Example**
-```bash
-zerogpu generate_followups \
-  "Solar panel adoption increased 35% in the US last year."
-```
-
-**Expected output (illustrative)**
-```json
-[
-  "Which states drove the largest share of the increase?",
-  "How does residential adoption compare to commercial?",
-  "What policy changes contributed to this growth?"
-]
-```
-
-**Outcomes** — same as the common table.
-
----
-
-### 4.15 `classify_domain`
+### 4.14 `classify_domain`
 
 Classify a domain name against the IAB taxonomy using `zlm-v1-iab-domain-classifier`. It infers the categories that characterize a site as a whole from the hostname alone — no crawl, no page text — which keeps payloads roughly 10x smaller than page-level classification. Use it for bidstream enrichment, allow/deny-list scoring, and inventory-level targeting; when you have the page text and need per-URL precision, use [`classify_iab`](#45-classify_iab) instead.
 
@@ -684,7 +648,7 @@ Content-Type: application/json
 x-api-key:    <ZEROGPU_API_KEY>
 ```
 
-The exceptions are `summarize`, `chat --model qwen3-30b-a3b-fp8`, `chat --model glm-5.2`, and `chat --model deepseek-v4-flash`, whose models the ZeroGPU platform serves only through the OpenAI-compatible Chat Completions endpoint:
+The exceptions are `summarize`, `chat --model qwen3-30b-a3b-fp8`, `chat --model glm-5.2`, and `chat --model deepseek-v4-flash-0731`, whose models the ZeroGPU platform serves only through the OpenAI-compatible Chat Completions endpoint:
 
 ```
 POST https://api.zerogpu.ai/v1/chat/completions

diff --git a/package-lock.json b/package-lock.json
diff --git a/package.json b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "zerogpu-cli",
-  "version": "3.7.2",
+  "version": "3.8.0",
   "description": "Command-line interface for ZeroGPU.",
   "type": "module",
   "bin": {

diff --git a/src/cli.ts b/src/cli.ts
@@ -14,7 +14,6 @@ import { registerCostSavingsCommand } from "./commands/costSavings.js";
 import { registerExtractEntitiesCommand } from "./commands/extractEntities.js";
 import { registerExtractJsonCommand } from "./commands/extractJson.js";
 import { registerExtractPiiCommand } from "./commands/extractPii.js";
-import { registerGenerateFollowupsCommand } from "./commands/generateFollowups.js";
 import { registerLoginCommand } from "./commands/login.js";
 import { registerRedactPiiCommand } from "./commands/redactPii.js";
 import { registerStatusCommand } from "./commands/status.js";
@@ -99,7 +98,6 @@ export function buildProgram(): Command {
   registerClassifyDomainCommand(program);
   registerClassifyStructuredCommand(program);
   registerClassifyZeroShotCommand(program);
-  registerGenerateFollowupsCommand(program);
   registerRedactPiiCommand(program);
   registerExtractPiiCommand(program);
   registerExtractEntitiesCommand(program);

diff --git a/src/commands/chat.ts b/src/commands/chat.ts
@@ -17,7 +17,7 @@ import { recordAndMaybeNotify } from "../lib/savings.js";
 const DEFAULT_MODEL = "LFM2.5-1.2B-Instruct";
 
 // Text-generation models `--model` accepts, and the API each one speaks.
-// qwen3-30b-a3b-fp8, glm-5.2, and deepseek-v4-flash are Chat Completions only —
+// qwen3-30b-a3b-fp8, glm-5.2, and deepseek-v4-flash-0731 are Chat Completions only —
 // they have no Responses endpoint.
 // Source: https://docs.zerogpu.ai/docs/text-generation
 const CHAT_MODELS: Record<string, "responses" | "chat-completions"> = {
@@ -26,7 +26,7 @@ const CHAT_MODELS: Record<string, "responses" | "chat-completions"> = {
   "gpt-oss-120b": "responses",
   "qwen3-30b-a3b-fp8": "chat-completions",
   "glm-5.2": "chat-completions",
-  "deepseek-v4-flash": "chat-completions",
+  "deepseek-v4-flash-0731": "chat-completions",
 };
 
 // Model ids are case-sensitive to the API but not to the person typing them.

diff --git a/src/commands/generateFollowups.ts b/src/commands/generateFollowups.ts
diff --git a/src/lib/savings.ts b/src/lib/savings.ts
@@ -23,24 +23,22 @@ export const ZGPU_PRICING: Record<string, { in: number; out: number }> = {
   "gpt-oss-120b": { in: 0.15, out: 0.6 },
   "qwen3-30b-a3b-fp8": { in: 0.05, out: 0.3 },
   "glm-5.2": { in: 1.1, out: 3.5 },
-  "deepseek-v4-flash": { in: 0.07, out: 0.14 },
+  "deepseek-v4-flash-0731": { in: 0.16, out: 0.38 },
   "llama-3.1-8b-instruct-fast": { in: 0.02, out: 0.05 },
   "zlm-v1-iab-classify-edge": { in: 0.02, out: 0.05 },
   "zlm-v2-iab-classify-edge-enriched": { in: 0.02, out: 0.05 },
   "zlm-v1-iab-domain-classifier": { in: 0.02, out: 0.05 },
-  "zlm-v1-followup-questions-edge": { in: 0.02, out: 0.05 },
+  "zlm-v1-signal-extract": { in: 0.02, out: 0.05 },
   "gliner-multi-pii-v1": { in: 0.02, out: 0.05 },
   "gliner2-base-v1": { in: 0.02, out: 0.05 },
   "deberta-v3-small": { in: 0.02, out: 0.05 },
   "LFM2.5-1.2B-Thinking": { in: 0.02, out: 0.05 },
   "LFM2.5-1.2B-Instruct": { in: 0.02, out: 0.05 },
   "zlm-v1-moderation-edge": { in: 0.02, out: 0.05 },
-  "t5-small": { in: 0.05, out: 0.4 },
-  "zlm-v1-multi-iab-classify": { in: 0.05, out: 0.4 },
   // Embedding models bill input tokens only; there are no output tokens to
   // charge, so `out: 0` is the real rate, not a placeholder.
-  "all-minilm-l6-v2": { in: 0.5, out: 0 },
-  "bge-small-en-v1.5": { in: 0.5, out: 0 },
+  "all-minilm-l6-v2": { in: 0.004, out: 0 },
+  "bge-small-en-v1.5": { in: 0.004, out: 0 },
 };
 // Conservative fallback for any model id not in the table above: the priciest
 // published rate, so an unlisted model never overstates savings. Currently

diff --git a/tests/savings.test.ts b/tests/savings.test.ts
@@ -93,7 +93,7 @@ describe("computeCallSavings", () => {
       "gpt-oss-120b",
       "qwen3-30b-a3b-fp8",
       "glm-5.2",
-      "deepseek-v4-flash",
+      "deepseek-v4-flash-0731",
       "LFM2.5-1.2B-Instruct",
     ]) {
       expect(unknown).toBeLessThanOrEqual(
@@ -110,31 +110,24 @@ describe("ZGPU_PRICING tracks the published model catalog", () => {
   // below then force src/lib/savings.ts to change with it. Without this guard the
   // CLI silently keeps a stale rate and misreports savings — which is exactly how
   // gpt-oss-120b sat at $0.03/$0.10 long after it was repriced to $0.15/$0.60.
-  //
-  // One deliberate exception: zlm-v1-followup-questions-edge was removed from
-  // the published catalog when the model was deprecated, but `zerogpu
-  // generate-followups` still routes to it, so the CLI must still price it.
-  // Drop both entries together when that command goes.
   const CATALOG: Record<string, { in: number; out: number }> = {
     "gpt-oss-120b": { in: 0.15, out: 0.6 },
     "qwen3-30b-a3b-fp8": { in: 0.05, out: 0.3 },
     "glm-5.2": { in: 1.1, out: 3.5 },
-    "deepseek-v4-flash": { in: 0.07, out: 0.14 },
+    "deepseek-v4-flash-0731": { in: 0.16, out: 0.38 },
     "llama-3.1-8b-instruct-fast": { in: 0.02, out: 0.05 },
     "zlm-v2-iab-classify-edge-enriched": { in: 0.02, out: 0.05 },
     "zlm-v1-iab-classify-edge": { in: 0.02, out: 0.05 },
     "zlm-v1-iab-domain-classifier": { in: 0.02, out: 0.05 },
-    "zlm-v1-followup-questions-edge": { in: 0.02, out: 0.05 },
+    "zlm-v1-signal-extract": { in: 0.02, out: 0.05 },
     "gliner-multi-pii-v1": { in: 0.02, out: 0.05 },
     "gliner2-base-v1": { in: 0.02, out: 0.05 },
     "deberta-v3-small": { in: 0.02, out: 0.05 },
     "LFM2.5-1.2B-Thinking": { in: 0.02, out: 0.05 },
     "LFM2.5-1.2B-Instruct": { in: 0.02, out: 0.05 },
     "zlm-v1-moderation-edge": { in: 0.02, out: 0.05 },
-    "t5-small": { in: 0.05, out: 0.4 },
-    "zlm-v1-multi-iab-classify": { in: 0.05, out: 0.4 },
-    "all-minilm-l6-v2": { in: 0.5, out: 0 },
-    "bge-small-en-v1.5": { in: 0.5, out: 0 },
+    "all-minilm-l6-v2": { in: 0.004, out: 0 },
+    "bge-small-en-v1.5": { in: 0.004, out: 0 },
   };
 
   it("prices every catalog model at the published rate", () => {