agents: revert --llm default to the newest preview, not the spec's 'default' (v0.3.2)

v0.3.1 followed Codex's advice to set --llm default to the OpenAPI
spec's nominal PromptAgentAPIModel.llm default (gemini-2.5-flash).
That's the older stable release. The spec's declared "default" string
lags by a couple of model generations. The ElevenLabs Agents backend
already accepts gemini-3-flash-preview and gemini-3.1-flash-lite-preview
in its LLM enum; our maintainer's live Guidance 40 session empirically
verified gemini-3.1-flash-lite-preview as the fastest reliable default.

Reverting to gemini-3.1-flash-lite-preview (the same default v0.2.x
shipped with). gemini-3.1-flash-live-preview is Google's specialized
voice model but isn't in the ElevenLabs LLM enum yet — we'll re-
evaluate once it is.

Touched: src/cli.rs (doc comment + default_value), src/help.rs (tips
copy + two examples), src/commands/agent_info.rs (defaults), src/
commands/agents/agent_config.rs (GOTCHAS), README.md, CHANGELOG.md.
Test suite still green (36 suites).

Author: longevityboris

CHANGELOG.md

@@ -3,6 +3,26 @@
 All notable changes to `elevenlabs-cli` are listed here.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versioning is [SemVer](https://semver.org).
 
+## [0.3.2] — 2026-04-21
+
+### Changed — `--llm` default reverted to the newest preview
+
+v0.3.1 moved the `agents create --llm` default to `gemini-2.5-flash`
+because the OpenAPI spec lists that string as the default on
+`PromptAgentAPIModel.llm`. That was a regression — the spec's "default"
+tracks the oldest-stable release, not the newest-preview. ElevenLabs
+already accepts several newer Gemini flash previews
+(`gemini-3-flash-preview`, `gemini-3.1-flash-lite-preview`); our
+maintainer's live Guidance 40 session empirically verified
+`gemini-3.1-flash-lite-preview` as the fastest reliable default. This
+release reverts to that.
+
+- `agents create --llm` default: `gemini-2.5-flash` →
+  `gemini-3.1-flash-lite-preview`. The specialized voice model
+  `gemini-3.1-flash-live-preview` exists at Google but isn't in the
+  ElevenLabs LLM enum yet; once it's added we'll re-evaluate.
+- Help / agent-info / GOTCHAS / README all aligned on the new default.
+
 ## [0.3.1] — 2026-04-21
 
 Post-ship review of v0.3.0 (Codex, gpt-5.4 xhigh) caught six real bugs +

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "elevenlabs"
-version = "0.3.1"
+version = "0.3.2"
 edition = "2024"
 rust-version = "1.85"
 description = "Agent-friendly CLI for the ElevenLabs AI audio platform (TTS, STT, voices, agents, music)"

Cargo.toml

@@ -272,7 +272,7 @@ elevenlabs agents show <agent_id>        # alias: get
 elevenlabs agents create <name>
     --system-prompt "..."
     [--first-message "Hi, how can I help?"]
-    [--voice-id ID] [--language en] [--llm gemini-2.5-flash]
+    [--voice-id ID] [--language en] [--llm gemini-3.1-flash-lite-preview]
     [--temperature 0.5] [--model-id eleven_flash_v2_5]
     [--expressive-mode] [--max-duration-seconds 600]
     [--voicemail-detection] [--voicemail-message "..."]

README.md

@@ -1175,12 +1175,18 @@ pub enum AgentsAction {
         #[arg(long, default_value = "en")]
         language: String,
 
-        /// LLM id. Default `gemini-2.5-flash` (matches the OpenAPI spec's
-        /// current default on `PromptAgentAPIModel.llm`). Discover the full
-        /// backend allowlist with `elevenlabs agents llms`. If
-        /// `conversations show` reports 0 output tokens on the chosen LLM,
-        /// the backend rejected it and fell back silently — swap LLMs.
-        #[arg(long, default_value = "gemini-2.5-flash")]
+        /// LLM id. Default `gemini-3.1-flash-lite-preview` — the newest
+        /// Gemini flash preview the ElevenLabs Agents backend accepts (the
+        /// OpenAPI spec's `PromptAgentAPIModel.llm` *default* string is
+        /// `gemini-2.5-flash`, but that's the older stable release; 3.1 is
+        /// current, empirically working, and lower-latency for turn-based
+        /// voice agents). `gemini-3.1-flash-live-preview` / `gemini-3.1-
+        /// flash-preview` exist at Google but aren't in the ElevenLabs
+        /// allowlist yet. Discover the full allowlist with
+        /// `elevenlabs agents llms`. If `conversations show` reports 0
+        /// output tokens on the chosen LLM, the backend rejected it and
+        /// fell back silently — swap LLMs.
+        #[arg(long, default_value = "gemini-3.1-flash-lite-preview")]
         llm: String,
 
         /// Temperature 0.0-1.0

src/cli.rs

@@ -153,7 +153,7 @@ pub fn run() {
                 "description": "Create a conversational AI agent. See `known_values.agent_tts_model_ids` and `gotchas.agents` before passing --model-id / --llm / --expressive-mode.",
                 "aliases": ["new"],
                 "defaults": {
-                    "--llm": "gemini-2.5-flash",
+                    "--llm": "gemini-3.1-flash-lite-preview",
                     "--model-id": "eleven_flash_v2_5",
                     "--temperature": 0.5,
                     "--language": "en",

src/commands/agent_info.rs

@@ -51,8 +51,10 @@ pub const GOTCHAS: &[&str] = &[
     "expressive_mode only takes effect with model_id=eleven_v3_conversational. With any other \
      model the server silently drops the flag — the PATCH succeeds but expressive_mode stays false.",
     "--llm accepts any string but the Agents backend has its own allowlist. Discover the live list \
-     with `elevenlabs agents llms`. The OpenAPI spec default is `gemini-2.5-flash`. If an accepted \
-     LLM fails to generate at conversation time (0 output tokens in `conversations show`), swap LLMs.",
+     with `elevenlabs agents llms`. This CLI defaults to `gemini-3.1-flash-lite-preview` (the newest \
+     Gemini flash preview in the ElevenLabs enum) — the OpenAPI spec default `gemini-2.5-flash` is \
+     the older stable release. If an accepted LLM fails to generate at conversation time (0 output \
+     tokens in `conversations show`), swap LLMs.",
     "Spec default for conversation.max_duration_seconds is 600 (10 min). This CLI's \
      `agents create` defaults to the same value. Bump to 1800 (30 min) or higher for long-form \
      interviews / coaching — calls hang up hard at this limit regardless of transcript state.",

src/commands/agents/agent_config.rs

@@ -248,12 +248,14 @@ EXAMPLES
 pub const AGENTS_CREATE_HELP: &str = "TIPS
  - --system-prompt is REQUIRED — there is no interactive fallback. Keep
    prompts specific; vague prompts yield vague agents.
- - Defaults: --llm gemini-2.5-flash (spec default on
-   PromptAgentAPIModel.llm), --model-id eleven_flash_v2_5 (lowest
-   TTS latency, current agent recommendation), --max-duration-seconds
-   600 (10 min, matching the spec default). Override --llm for
-   reasoning-heavy use cases; override --model-id for higher-fidelity
-   voice; bump --max-duration-seconds for long-form interviews.
+ - Defaults: --llm gemini-3.1-flash-lite-preview (newest Gemini flash
+   preview the ElevenLabs backend accepts — the OpenAPI spec nominally
+   defaults to the older gemini-2.5-flash but 3.1 is current and lower
+   latency in practice), --model-id eleven_flash_v2_5 (lowest TTS
+   latency, current agent recommendation), --max-duration-seconds 600
+   (10 min, matching the spec default). Override --llm for reasoning-
+   heavy use cases; override --model-id for higher-fidelity voice;
+   bump --max-duration-seconds for long-form interviews.
  - Valid --model-id values (server-enforced allowlist):
    eleven_flash_v2_5 (recommended), eleven_flash_v2, eleven_multilingual_v2,
    eleven_v3_conversational, eleven_turbo_v2_5 (DEPRECATED —
@@ -299,7 +301,7 @@ EXAMPLES
  $ elevenlabs agents create \"Research Assistant\" \\
      --system-prompt \"$(cat prompts/ra.txt)\" \\
      --voice-id 21m00Tcm4TlvDq8ikWAM \\
-     --llm gemini-2.5-flash --model-id eleven_multilingual_v2
+     --llm gemini-3.1-flash-lite-preview --model-id eleven_multilingual_v2
 
  # Attach docs after create
  $ AGENT=$(elevenlabs agents create ... --json | jq -r '.data.agent_id')
@@ -365,7 +367,7 @@ EXAMPLES
  $ elevenlabs agents update agent_abc --patch p.json
 
  # Change just the LLM
- $ echo '{\"conversation_config\":{\"agent\":{\"prompt\":{\"llm\":\"gemini-2.5-flash\"}}}}' > p.json
+ $ echo '{\"conversation_config\":{\"agent\":{\"prompt\":{\"llm\":\"gemini-3.1-flash-lite-preview\"}}}}' > p.json
  $ elevenlabs agents update agent_abc --patch p.json
 
  # Rename the agent (top-level, not inside conversation_config)