Platform: Electron
TaskWraith's model picker is provider-aware: choose a provider, then a model, the reasoning level it supports, and (where offered) a Fast tier. This page is the concise, public reference for the curated picker catalogue.
Snapshot: 29 August 2026. Your actual picker is still governed by the provider CLI, your account and plan, and (for Ollama) the models installed on your machine. Codex is refreshed from its live provider catalogue when available; the rows below describe TaskWraith's curated fallback and the standard options it presents. Rows added or corrected after v1.9.6 describe the current source-ahead checkout and are not a v1.9.6 release guarantee.
- Default marks TaskWraith's initial selection for a new provider seat.
- Reasoning uses the composer's common ladder: Light, Medium, High, Extra, Max, and Ultracode. A dash means that model has no configurable TaskWraith reasoning control.
- Fast is provider-specific: Toggle exposes the Fast control in the picker, Included means the provider's model is always Fast, and Pair means normal and Fast are separate model rows. Fast availability and pricing depend on the provider plan.
The thin colour rail beside each provider is the same provider hue used by the in-app New Additions notification. The marks are the actual first-party provider logos from the project's provider-logo catalogue, not the monoline glyph set.
|
Codex / OpenAI Violet provider hue · active for new runs |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
GPT-5.6-Sol gpt-5.6-sol |
Light · Medium · High · Extra · Max · Ultracode | Toggle | Latest frontier agentic coding model. |
GPT-5.6-Terra gpt-5.6-terra |
Light · Medium · High · Extra · Max · Ultracode | Toggle | Balanced agentic coding for everyday work. |
GPT-5.6-Luna gpt-5.6-luna |
Light · Medium · High · Extra · Max | Toggle | Fast and affordable agentic coding. |
GPT-5.5 gpt-5.5 (Default) |
Light · Medium · High · Extra | Toggle | Default while the GPT-5.6 rollout remains account-dependent. |
GPT-5.4 gpt-5.4 |
Light · Medium · High · Extra | Toggle | Still offered when directly runnable even if discovery omits it. |
GPT-5.4 Mini gpt-5.4-mini |
Light · Medium · High · Extra | — | Smaller GPT-5.4 option. |
GPT-5.3 Codex Spark gpt-5.3-codex-spark |
Light · Medium · High · Extra | — | Research preview where available. |
|
Claude / Anthropic Clay provider hue · active for new runs |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Opus 5 claude-opus-5 |
Light · Medium · High · Extra · Max · Ultracode | Toggle | 1M context by default, adaptive thinking. |
Fable 5.1 claude-fable-5-1 |
Light · Medium · High · Extra · Max · Ultracode | — | 1M context, adaptive thinking. |
Sonnet 5 claude-sonnet-5 (Default) |
Light · Medium · High · Extra · Max · Ultracode | — | 1M context, extended thinking. |
Fable 5 Legacy claude-fable-5 |
Light · Medium · High · Extra · Max · Ultracode | — | 1M context legacy Fable, adaptive thinking. |
Sonnet 4.6 Legacy claude-sonnet-4-6 |
Light · Medium · High · Max | — | 200K context legacy Sonnet. |
Opus 4.8 1M Legacy claude-opus-4-8-1m |
Light · Medium · High · Extra · Max · Ultracode | Toggle | 1M context legacy Opus, extended thinking. |
Opus 4.7 1M Legacy claude-opus-4-7-1m |
Light · Medium · High · Extra · Max · Ultracode | Toggle | 1M context legacy Opus. |
Haiku 4.5 claude-haiku-4-5 |
— | — | Fast and efficient; no configurable reasoning control. |
|
Kimi / Moonshot Blue provider hue · catalogued for admitted runtimes; packaged source-ahead builds await a commissioned tuple |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
K2.7 Coding kimi-k2.7-code (Default) |
On (fixed) | Toggle — Highspeed | Thinking is always on. |
K3 (up to 1M) kimi-k3 |
Low · High · Max | — | 256K on Moderato; up to 1M on Allegretto+ tiers. |
K3 256K kimi-k3-256k |
Low · High · Max | — | Quota-efficient K3 route — fixed 256K context. Never inherits K2.7's Highspeed tier. |
|
Grok / xAI Neutral provider hue · active for new runs |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Grok 4.6 Fast grok-4.6 (Default) |
Low · Medium · High · Extra-high | Included | 500K-context coding model; Fast is provider-encoded. |
Grok 4.5 Fast grok-4.5 |
Low · Medium · High | Included | Previous default; retained as a selectable row. |
Grok Composer 2.5 Fast grok-composer-2.5-fast |
— | Included | Historical/specialised Fast model row. |
|
Cursor Mustard provider hue · managed runs with native and TaskWraith tools |
Managed Cursor is selectable for new runs and supports both Cursor's native tools and TaskWraith's governed tool gateway. The selected permission posture and workspace Tool Grants apply to TaskWraith-mediated calls; see Trust & Safety for provider-native boundaries.
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Composer 2.5 Fast composer-2.5-fast (Default) |
— | Included | Fast route is encoded in the selected model row. |
Composer 2.5 composer-2.5 |
— | Pair | Select this normal row or its Fast counterpart above. |
Cursor Grok 4.6 grok-4.6 |
Low · Medium · High · Extra-high | Toggle | First-party Cursor model pool with 256K context. |
Cursor Grok 4.5 grok-4.5 |
Low · Medium · High | Toggle | First-party Cursor model pool with 500K context. |
|
AntiGravity / Google Green provider hue · consented official CLI + separately billed Gemini API-key lanes |
The separately consented official agy CLI lane follows the signed-in
AntiGravity subscription catalogue. Its newest family is:
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Gemini 3.7 Flash gemini-3.7-flash-{low,medium,high} |
Low · Medium · High | — | Authenticated agy models catalogue; availability remains account-controlled. |
The independent bring-your-own-key lane curates the authenticated Gemini API
models.list response, so the live picker—not a frozen documentation table—is
authoritative. Its current static fallback floor is:
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Gemini 3.6 Flash gemini-api:gemini-3.6-flash |
— | — | Latest balanced model for agentic and multimodal work. |
Gemini 3.5 Flash gemini-api:gemini-3.5-flash |
— | — | Sustained frontier performance for agentic and coding work. |
Gemini 3.1 Pro Preview gemini-api:gemini-3.1-pro-preview |
— | — | Preview-tier reasoning model retained in the fallback floor. |
Gemini 3.1 Flash-Lite gemini-api:gemini-3.1-flash-lite |
— | — | Lowest-cost row for high-throughput execution. |
Live discovery wins for both lanes. The consent-gated CLI catalogue has a
current mixed-family floor so a transient agy models failure does not hide
the provider; the API-key lane likewise retains bounded fallback rows, and the
four rows above are exactly that floor — the 3.6, 3.5, and 3.1 families. The
2.5 family was dropped from the floor because gemini-2.5-flash now 404s as
"no longer available to new users". A stale row still fails visibly at
dispatch. See Google's
Gemini API model catalogue for
the upstream API lifecycle.
| Muse Code / Meta Meta blue provider hue · active for new runs |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Muse Spark 1.2 muse-spark-1.2 (Default) |
Light · Medium · High · Extra · Ultracode | — | 200K context · $1.25/$4.25 per Mtok. |
Muse's model lifecycle is provider-published from the local Muse CLI catalogue, so the live picker follows the CLI's own current-model set; the row above is TaskWraith's curated fallback. Muse's own picker labels its stops Minimal · Low · Medium · High · Extra High · Ultra; the Reasoning column above maps them onto the shared ladder vocabulary.
|
Mistral Vibe Mistral orange provider hue · active for new runs |
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
Devstral Small devstral-small (Default) |
— | — | 256K context · coding-tuned · $0.10/$0.30 per Mtok. |
Mistral Medium 3.5 mistral-medium-3.5 |
High (fixed) | — | 256K context · flagship · $1.50/$7.50 per Mtok. Always thinks at High. |
Mistral Large 3 mistral-large-2512 |
— | — | 262K context · flagship · $0.50/$1.50 per Mtok. |
GLM-5.2 (via Mistral) zai-glm-5-2 |
— | — | 1M context · $1.40/$4.40 per Mtok. |
Codestral (Aug 2025) codestral-2508 |
— | — | 131K context · coding-tuned · $0.30/$0.90 per Mtok. |
Mistral Small 4 mistral-small-2603 |
— | — | 256K context · $0.15/$0.60 per Mtok. |
Devstral 2 devstral-2512 |
— | — | 262K context · $0.40/$2.00 per Mtok. |
Leanstral 1.5 (Labs) labs-leanstral-1-5 |
— | — | 262K context · free research tier. |
Mistral Medium (Latest) mistral-medium-latest |
— | — | 262K context · flagship · $1.50/$7.50 per Mtok. |
Mistral Medium 3.1 mistral-medium-2508 |
— | — | 262K context · $0.40/$2.00 per Mtok. |
Mistral Medium 3 mistral-medium-2505 |
— | — | 131K context · $0.40/$2.00 per Mtok. |
Ministral 3 (14B) ministral-14b-2512 |
— | — | 262K context · $0.20/$0.20 per Mtok. |
Ministral 3 (8B) ministral-8b-2512 |
— | — | 262K context · $0.15/$0.15 per Mtok. |
Ministral 3 (3B) ministral-3b-2512 |
— | — | 262K context · $0.10/$0.10 per Mtok. |
Only the first two rows bill against the Vibe subscription. Every row below
them is API-only: it runs on a user-supplied Mistral API key
(MISTRAL_API_KEY), metered per token and separate from any Vibe plan.
The picker marks those rows with the API-key glyph so the lane is visible
before you start a run.
Vibe's third catalogue entry, local, is a llamacpp backend and is
deliberately omitted: local inference is Ollama's lane here, and listing it
would leave a permanently dead row for anyone without their own llama-server.
|
Devin / Cognition Steel-blue provider hue · active for new runs |
| Model | Vendor | Reasoning (default in bold) | List price (in / out per 1M tokens) | Notes |
|---|---|---|---|---|
SWE-1.6 Slow swe-1-6-slow (Default) |
Cognition | — | $0.5 / $2.5 | Seat default; the model a fresh Devin CLI install pins in its own config |
SWE-1.6 swe-1-6 |
Cognition | — | $0.5 / $2.5 | — |
SWE-1.6 Fast swe-1-6-fast |
Cognition | — | $0.5 / $2.5 | — |
SWE-1.7 swe-1-7 |
Cognition | Medium · Max | $0.5 / $2.5 | The family the CLI resolves a bare swe-1.7 to |
SWE-1.7 Lightning swe-1-7-lightning |
Cognition | Medium · Max | $2.5 / $12.5 | swe alias resolves here |
Adaptive adaptive |
Cognition | — | $0.5 / $2 | Cognition's model router; enterprise admins must enable it |
Claude Fable 5.1 claude-fable-5-1 |
Anthropic | Low · Medium · High · Extra High · Max | $10 / $50 | new |
Claude Opus 5 claude-opus-5 |
Anthropic | Low · Medium · High · Extra High · Max | $5 / $25 | opus alias resolves here |
Claude Sonnet 5 claude-sonnet-5 |
Anthropic | Low · Medium · High · Extra High · Max | $2 / $10 | claude / sonnet alias resolves here |
GPT-5.6 Sol gpt-5-6-sol |
OpenAI | None · Low · Medium · High · Extra High · Max | $4 / $20 | — |
GPT-5.6 Terra gpt-5-6-terra |
OpenAI | None · Low · Medium · High · Extra High · Max | $2 / $12 | gpt alias resolves here |
GPT-5.6 Luna gpt-5-6-luna |
OpenAI | None · Low · Medium · High · Extra High · Max | $0.2 / $1.2 | — |
GPT-5.3-Codex gpt-5-3-codex |
OpenAI | Low · Medium · High · Extra High | $1.75 / $14 | codex alias resolves here |
Gemini 3.7 Flash gemini-3-7-flash |
Low · Medium · High | $1.5 / $7.5 | gemini alias resolves here |
|
Grok 4.6 grok-4-6 |
xAI | Low · Medium · High · Extra High | $2 / $6 | new; beta |
Grok 4.5 grok-4-5 |
xAI | Low · Medium · High | $2 / $6 | — |
Kimi K3 kimi-k3 |
Moonshot AI | Low · High · Max | $3 / $15 | — |
GLM-5.3 glm-5-3 |
Z.ai | Low · High · Max | $1.4 / $4.4 | — |
GLM-5.3 Flash glm-5-3-flash |
Z.ai | Low · High · Max | $0.15 / $0.5 | — |
GLM-5.2 glm-5-2 |
Z.ai | None · High · Max | $1.4 / $4.4 | — |
GLM-5.2 1M glm-5-2-1m |
Z.ai | None · High · Max | $0.7 / $2.2 | 1M-context variants of GLM-5.2 |
DeepSeek V4 Pro deepseek-v4-pro |
DeepSeek | Low · High · Max | $1.32 / $3.96 | — |
DeepSeek V4 Flash deepseek-v4-flash |
DeepSeek | Low · High · Max | $0.14 / $0.28 | — |
Inkling inkling |
Thinking Machines | None · Low · Medium · High · Extra High · Max | $1.4 / $4.4 | — |
Nemotron 3 Ultra nemotron-3-ultra |
NVIDIA | None · Medium · High | $0.6 / $2.4 | — |
Devin's rows are the model families the Devin CLI itself enumerates
(devin models list --format json, CLI 3000.6.7, retrieved 2026-09-01), one
picker row per family with the CLI's own label and the list price it showed for
a self-serve seat; enterprise seats are metered in ACUs instead. A family's
reasoning column is its variant ladder: the ordinary effort slider picks the
level and the run folds it into the exact CLI variant as
devin acp --model <family>-<level> (SWE-1.7 at Max is the bare swe-1-7
uid), so "Claude Opus 5" plus "High" dispatches claude-opus-5-high. The
picker carries Cognition's own families and the Adaptive router in full, plus
the newest generation of every other vendor line without the double-price
-fast / -priority speed tiers; the curation rule lives in
src/shared/devinModelCatalog.ts. There is no "CLI default" row — a legacy
stored cli-default selection resolves to SWE-1.6 Slow, and a custom id
outside the catalogue passes through verbatim and fails visibly at the CLI if
unknown. The seat is the Devin CLI over ACP on your own paid seat. It
authenticates through WINDSURF_API_KEY / DEVIN_API_KEY or the credentials
file written by devin auth login, and Settings → Providers → Devin can pin a
custom API server URL (HTTPS only; HTTP on loopback).
|
Pi Slate provider hue · active for new runs · bring your own upstream keys |
Pi brokers several upstreams behind one seat. Unlike AntiGravity's API lane, the catalogue is not live-discovered: it is a version-pinned list extracted from the Pi CLI's bundled catalogue, because a provider whose model list can transiently come back empty would vanish from every picker. You only see rows for upstreams you have configured a key for, so this table is broader than what any one install shows. Rows retire on dated lifecycle entries.
Two upstreams are folded in the ids below. The Xiaomi token plan ships three
regional deployments of one identical MiMo catalogue
(xiaomi-token-plan-{cn,sgp,ams}); Settings → Providers → Pi files your key
under exactly one region, so only that region's rows are ever offered. The
OpenRouter rows are user-approved exceptions rather than a general OpenRouter
lane, and their metadata is written into the run's isolated Pi home at launch.
| Model | Reasoning | Fast | Notes |
|---|---|---|---|
DeepSeek V4 Pro deepseek/deepseek-v4-pro |
— | — | 1M context via DeepSeek. |
DeepSeek V4 Flash deepseek/deepseek-v4-flash (Default) |
— | — | 1M context via DeepSeek. |
GLM-5.2 zai/glm-5.2 |
— | — | 1M context via Z.ai. |
GLM-5.1 zai/glm-5.1 |
— | — | 200K context via Z.ai. |
GLM-4.7 zai/glm-4.7 |
— | — | ~200K context via Z.ai. |
Qwen3.7 Max qwen-token-plan/qwen3.7-max |
— | — | 1M context via the Qwen token plan. |
Qwen3.7 Plus qwen-token-plan/qwen3.7-plus |
— | — | 1M context via the Qwen token plan. |
Qwen3.8 Max qwen-token-plan/qwen3.8-max |
— | — | 1M context via the Qwen token plan. The …-max-preview id survives only as a legacy alias for older transcripts. |
MiniMax M3 minimax/MiniMax-M3 |
Off / High | — | 1M context via MiniMax; High is the on/adaptive control. |
MiniMax M2.7 minimax/MiniMax-M2.7 |
— | — | ~200K context via MiniMax. |
MiMo V2 Pro xiaomi-token-plan-{cn,sgp,ams}/mimo-v2-pro |
— | — | 1M context via the Xiaomi token plan. Retired 2026-08-30 after Xiaomi sunset it in favor of MiMo V2.5 and MiMo V2.5 Pro; no longer offered for new runs, kept here so older transcripts still decode. |
MiMo V2.5 xiaomi-token-plan-{cn,sgp,ams}/mimo-v2.5 |
— | — | 1M context via the Xiaomi token plan. |
MiMo V2.5 Pro xiaomi-token-plan-{cn,sgp,ams}/mimo-v2.5-pro |
— | — | 1M context via the Xiaomi token plan. |
Devstral 2512 mistral/devstral-2512 |
— | — | 256K context via Mistral. |
Mistral Medium 3.5 mistral/mistral-medium-3.5 |
High (fixed) | — | 256K context via Mistral. Always thinks at High. |
Mistral Large 3 (2512) mistral/mistral-large-2512 |
— | — | 256K context via Mistral. Not a reasoning model. |
GLM-5.2 (via Mistral) mistral/zai-glm-5-2 |
— | — | 1M context via Mistral. |
Mistral Medium (Latest) mistral/mistral-medium-latest |
— | — | 256K context via Mistral. |
Mistral Small 4 mistral/mistral-small-2603 |
— | — | 256K context via Mistral. |
Mistral Medium 3.1 mistral/mistral-medium-2508 |
— | — | 256K context via Mistral. |
Mistral Medium 3 mistral/mistral-medium-2505 |
— | — | 131K context via Mistral. |
Codestral (Aug 2025) mistral/codestral-2508 |
— | — | 131K context via Mistral. |
Leanstral 1.5 (Labs) mistral/labs-leanstral-1-5 |
— | — | 256K context via Mistral. Free research tier. |
Ministral 3 (14B) mistral/ministral-14b-2512 |
— | — | 256K context via Mistral. |
Ministral 3 (8B) mistral/ministral-8b-2512 |
— | — | 256K context via Mistral. |
Ministral 3 (3B) mistral/ministral-3b-2512 |
— | — | 256K context via Mistral. |
GPT-OSS 120B (Groq) groq/openai/gpt-oss-120b |
— | — | 131K context via Groq. |
Qwen3 32B (Groq) groq/qwen/qwen3-32b |
— | — | 131K context via Groq. |
GLM-4.7 (Cerebras) cerebras/zai-glm-4.7 |
— | — | 131K context via Cerebras. Retired 2026-08-17; no longer offered for new runs, kept here so older transcripts still decode. |
GPT-OSS 120B (Cerebras) cerebras/gpt-oss-120b |
— | — | 131K context via Cerebras. |
Ox Alpha openrouter/stealth/ox-alpha |
— | — | Retired 2026-08-28 after OpenRouter withdrew it; no longer offered for new runs, kept here so older chats and saved ensemble seats still decode. |
GLM 5.2 openrouter/z-ai/glm-5.2 |
— | — | 256K context via OpenRouter. |
Laguna S 2.1 openrouter/poolside/laguna-s-2.1 |
— | — | 256K context via OpenRouter. |
Nemotron 3 Ultra openrouter/nvidia/nemotron-3-ultra-550b-a55b:free |
— | — | 1M context via OpenRouter. |
North Mini Code (OpenRouter Free) openrouter/cohere/north-mini-code:free |
Off / High | — | 256K context, 64K output, text and tools via OpenRouter. |
MiniMax M3 (OpenRouter Free) openrouter/minimax/minimax-m3:free |
Off / High | — | 1M multimodal context via OpenRouter. |
Inkling (OpenRouter Free) openrouter/thinkingmachines/inkling:free |
Off → Max | — | 1M text/image context, 256K output; Minimal, Low, Medium, High, and Max (no Extra High). |
Inkling Small (OpenRouter Free) openrouter/thinkingmachines/inkling-small:free |
Off → Max | — | Faster Inkling route with the same 1M context and effort ladder. |
Pi sends --thinking only when a reasoning choice is set, and every picker is
filtered through the selected route's own ladder. Boolean routes expose only
Off and High; always-on routes hide Off; Inkling keeps Off and Minimal distinct
and omits unsupported Extra High. UltraTask/top-tier requests clamp to the
selected model's real ceiling instead of forwarding a level that route ignores.
Leaving the control unset preserves the upstream default.
The two free Inkling endpoints are research, agentic-harness-only routes. Thinking Machines logs prompts and outputs for model improvement, so TaskWraith surfaces them without gating the user-approved capability and warns users not to send confidential or personal data through those free endpoints.
|
Ollama / local Green provider hue · active when Ollama is installed and reachable |
Ollama now has two lanes, and the picker separates them. Local models are discovered from the local daemon, so only pulled model tags can run; the curated presets below make common choices easy to find, and Custom model ID accepts another compatible installed tag.
Cloud models are discovered from your signed-in Ollama account rather than from pulled tags, so they run without being present on disk (shipped in v1.9.6). Discovery reports whether the daemon supports cloud, whether cloud is enabled, and whether you are authenticated; a model may additionally name a required plan. Nothing appears unless the daemon both supports and enables cloud, so a local-only install is unchanged. In the composer picker a cloud group is marked with a cloud icon in place of the usual provider swatch, and model preflight states cloud rather than local expectations for those rows.
There is no generic TaskWraith reasoning or Fast switch for either lane—native tool, vision, and thinking capabilities remain model-specific.
| Curated local model | Model ID | Picker note |
|---|---|---|
| Qwen 3 (4B Param) (Default) | qwen3:4b-instruct |
262K local context. |
| Qwen 3.5 (2B Param) | qwen3.5:2b |
262K; vision, tools, thinking. |
| Qwen 3.5 (4B Param) | qwen3.5:4b |
262K local context. |
| Qwen 3.5 (9B) | qwen3.5:9b |
262K local context. |
| Qwen 3.6 (35B-A3B) | qwen3.6:35b |
262K; vision, tools, thinking. |
| Qwen 3.8 (27B-MLX) | qwen3.8:27b-mlx |
262K; vision, tools, thinking. |
| Qwen 3.8 Flash Next (125B-MLX) | qwen3.8-flash-next:125b-mlx |
262K; vision, tools, thinking; 6B active. |
| Gemma 3 (4B Param) | gemma3:4b |
131K; vision. |
| Gemma 4 (12B) | gemma4:12b |
262K local context. |
| Gemma 4 (31B-MLX) | gemma4:31b-mlx |
262K local context. |
| Ornith 1.0 (9B) | ornith:9b |
262K agentic coding. |
| Ornith 1.0 (35B) | ornith:35b |
262K agentic coding. |
| Ornith 1.5 (9B Param) | ornith-1.5:9b |
262K agentic coding. |
| Ornith 1.5 (35B Param) | ornith-1.5:35b |
262K agentic coding. |
| Laguna XS 2.1 (33B-A3B Q8) | laguna-xs-2.1:q8_0 |
262K; tools and thinking. |
| GPT OSS (20B) | gpt-oss:20b |
131K local context. |
| LFM 2.5 Thinking (1.2B Param) | lfm2.5-thinking:1.2b |
128K; tools, thinking. |
| LFM 2.5 (8B-A1B) | lfm2.5:8b |
128K; tools, thinking. |
| MiniCPM-V 4.5 (8B) | minicpm-v4.5:8b |
40K; vision, tools, thinking. |
| Granite 4.0 (3B Param) | granite4:3b |
131K; tools. |
| Granite 4.1 (3B) | granite4.1:3b |
131K; tools. |
| Granite 4.1 (30B) | granite4.1:30b |
131K; tools. |
| Granite 4.2 (3B) | granite4.2:3b |
131K; tools, thinking. |
| Granite 4.2 (8B) | granite4.2:8b |
131K; tools, thinking. |
| Granite 4.2 (30B) | granite4.2:30b |
131K; tools, thinking. |
| Nemotron 3 Nano (4B Param) | nemotron-3-nano:4b |
262K; tools, thinking. |
| Nemotron 3 Nano Omni (33B) | nemotron3:33b |
131K; vision, tools, thinking. |
| Nemotron 3.5 Lightning (30B-MLX) | nemotron-3.5-lightning:30b-mlx |
262K; tools, thinking; 3B active. |
| Devstral Small 2 (24B Param) | devstral-small-2:24b |
393K; vision, tools; agentic coding. |
| Mistral Medium 3.5 (128B Param) | mistral-medium-3.5:128b |
262K; vision, tools, thinking; agentic coding. |
| Ministral 3 (3B Param) | ministral-3:3b |
262K; vision, tools. |
| Ministral 3 (14B Param) | ministral-3:14b |
262K; vision, tools. |
| Muse Glimmer (30B-MLX) | muse-glimmer:30b-mlx |
131K; vision, tools, thinking. |
| Llama 3.1 (8B Param) | llama3.1:8b |
131K; tools. |
| DeepSeek R1 (1.5B Param) | deepseek-r1:1.5b |
131K; tools, thinking. |
| DeepSeek R1 (8B Param) | deepseek-r1:8b |
131K; tools, thinking. |
| Rnj-1 (8B Param) | rnj-1 |
33K; tools; agentic coding. |
| GLM-4.7-Flash (30B-A3B Q4) | glm-4.7-flash:q4_K_M |
203K; tools, thinking. |
| North Mini Code 1.0 (30B-A3B Q4) | north-mini-code-1.0:q4_K_M |
500K; tools, thinking. |
| Llama 3.2 (3B Param) | llama3.2:3b |
131K; tools. |
| Custom model ID | custom |
Enter a compatible locally installed model tag. |
|
Gemini / Google (legacy provider id) Blue provider hue · historical chats remain readable; new runs are retired |
Historical records under the older standalone gemini provider id can contain
Auto (auto), Pro (pro), Flash (flash), and Flash Lite
(flash-lite, the previous default). They are shown for history and usage
continuity only, not as new-run choices; the live AntiGravity integration above
is a separate provider path.
The live picker remains authoritative. When supported models or their controls change, update this page alongside StaticProviderModels.ts, the picker defaults, and the Provider, Model, and Permissions Pickers guide. For official context-window and rate references, use Settings → Data → Model usage.
Provider marks remain the property of their respective owners and are shown only to identify supported integrations. See the linked provider-logo catalogue for source and trademark guidance.