ModelBench
Arcee Ai logo
Models/Arcee Ai

Arcee Ai: Trinity Large Thinking

Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use

Source-linked trinity Open weights OpenMDW-1.1 Released 2026-04-01
API record Report
InputT
OutputT
Input price$0.25/M
Output price$0.9/M
Context524.288K
Max output262.144K
Providers5
Inference availability

Providers

Provider-specific identifiers, limits, and listed prices per million tokens. Every row links back to the provider's own documentation.

Report a price
Providers offering Trinity Large Thinking
ProviderProvider model IDContextMax outputInputOutputCache readCapabilitiesDocs
Arcee trinity-large-thinking 262.144K 262.144K $0.25 $0.8 $0.06 ReasoningTools Docs ↗
Vercel AI Gateway arcee-ai/trinity-large-thinking 262.1K 80K $0.25 $0.9 — ReasoningTools Docs ↗
Kilo Gateway arcee-ai/trinity-large-thinking 262.144K 80K $0.25 $0.8 $0.06 ReasoningTools Docs ↗
OpenRouter arcee-ai/trinity-large-thinking 262.144K 80K $0.25 $0.8 $0.06 ReasoningTools Docs ↗
NanoGPT arcee-ai/trinity-large-thinking 262.144K 80K $0.25 $0.9 $0.125 ReasoningTools Docs ↗

Capability badges appear only where the provider catalog explicitly lists support. A blank cell means the source is silent, not that the feature is absent.

Listed rates

Price across providers

Input price per million tokens as published by each provider. Bars are drawn from listed rates only — no traffic weighting, since the catalog observes no requests.

Lowest input $0.25/M

Across 5 priced providers

Median input $0.25/M

Midpoint of listed rates

Highest input $0.25/M

Same as the lowest listed rate

Output range $0.8 – $0.9

Per million output tokens

Arcee $0.25/M
Vercel AI Gateway $0.25/M
Kilo Gateway $0.25/M
OpenRouter $0.25/M
NanoGPT $0.25/M
Cost calculator

Estimate a workload

$0.00
Excludes taxes, non-token charges, and tiered discounts.

5 providers list the identical $0.25 input rate, so price alone will not separate them — compare context limits, max output, and capabilities above.

Context limits also differ by provider, from 262.1K to 262.144K tokens. Compare the provider table above before choosing on price alone.

Specification

Capabilities

Recorded from the source catalog and provider listings.

✓ Reasoning Yes
✓ Tool calling Yes
× Structured output No
× Attachments No
× Vision input No
✓ Open weights Yes
Creator
Arcee Ai
Model family
trinity
Knowledge cutoff
Not documented
License
OpenMDW-1.1
Release date
2026-04-01
Model ID
arcee-ai/trinity-large-thinking

Weights: Hugging Face ↗

Published evaluations

Benchmarks

Every result stays attached to its source, version, metric and harness. Scores from different versions are never merged, and the profile below plots each benchmark against its own population rather than on a shared scale.

Benchmark registry
Benchmark profile
Coding IndexCoding Index Agentic IndexAgentic Index IntelligenceIntelligence Index Design Arena: w…Design Arena: website Design Arena: c…Design Arena: codecategories Design Arena: g…Design Arena: gamedev Design Arena: d…Design Arena: dataviz Design Arena: u…Design Arena: uicomponent This model — Coding Index: 25.8 index (26.1th percentile) This model — Agentic Index: 1.2 index (12.2th percentile) This model — Intelligence Index: 10.9 index (13.8th percentile) This model — Design Arena: website: 1149 elo (33.4th percentile) This model — Design Arena: codecategories: 1132 elo (27.2th percentile) This model — Design Arena: gamedev: 1102 elo (23.2th percentile) This model — Design Arena: dataviz: 1114 elo (21.5th percentile) This model — Design Arena: uicomponent: 1060 elo (17.1th percentile)

Each axis is this model's percentile among the 8 benchmarks it has published results for, measured against every other model with a score on that same benchmark. Percentiles are used because benchmarks do not share a scale — a 60 on one is not a 60 on another. Hover any point for the raw score.

Coding Index index Source ↗
Coding Index #152 of 205
Agentic Index index Source ↗
Agentic Index #169 of 193
Intelligence Index index Source ↗
Intelligence Index #166 of 192
Design Arena: website elo Source ↗
Design Arena: website #117 of 175
Design Arena: codecategories elo Source ↗
Design Arena: codecategories #122 of 167
Design Arena: gamedev elo Source ↗
Design Arena: gamedev #127 of 166
Design Arena: dataviz elo Source ↗
Design Arena: dataviz #130 of 165
Design Arena: uicomponent elo Source ↗
Design Arena: uicomponent #134 of 161
Design Arena: 3d elo Source ↗
Design Arena: 3d #110 of 157
Design Arena: svg elo Source ↗
Design Arena: svg #107 of 116
Design Arena: asciiart elo Source ↗
Design Arena: asciiart #96 of 99

Each strip shows every published score for that benchmark, with this model marked. The lighter shape behind the ticks is the density of results, and the dashed line is the median.

Cost against capability

Price and performance

Listed input price plotted against Coding Index, the benchmark with the widest published coverage that this model appears in.

0 43 86 $0.01 $0.1 $1 $10 Claude Opus 5 — $5/M, score 78.0 OpenAI: GPT-5.1 (batch) — $0.625/M, score 49.4 Anthropic: Claude Haiku 4.5 (batch) — $0.5/M, score 43.9 OpenAI: GPT-5 (batch) — $0.625/M, score 37.8 Gemini 3.1 Flash Lite Preview — $0.25/M, score 34.7 Gemma 3 27B IT — $0.08/M, score 10.1 OpenAI: GPT-4.1 Mini (batch) — $0.2/M, score 20.2 Gemma 4 31B IT — $0.09/M, score 43.4 Mistral: Ministral 3 8B 2512 (batch) — $0.075/M, score 9.7 GPT-5.4 — $2.5/M, score 71.1 GPT-5.6 Luna — $0.2/M, score 71.4 Gemini 3.5 Flash Lite — $0.3/M, score 49.3 Mistral: Mistral Large 3 2512 (batch) — $0.25/M, score 20.1 OpenAI: o3 Mini High (batch) — $0.55/M, score 16.3 Qwen: Qwen3.8 Max (0902) — $2/M, score 71.8 Gemini 3.5 Flash — $1.5/M, score 70.1 Z.ai: GLM 5.3 Flash (batch) — $0.075/M, score 71.5 Gemma 3 12B IT — $0.05/M, score 5.8 GPT-4.1 mini — $0.4/M, score 20.2 DeepSeek V4 Pro — $0.435/M, score 59.4 Claude Sonnet 5 — $2/M, score 71.5 Mistral: Ministral 3 8B 2512 — $0.15/M, score 9.7 OpenAI: o1 (batch) — $7.5/M, score 39.7 Z.ai: GLM 5.1 — $0.966/M, score 55.8 Gemini 3.7 Flash — $0.75/M, score 76.1 LongCat-2.0 — $0.3/M, score 45.3 Anthropic: Claude Opus 4.8 (batch) — $2.5/M, score 74.3 Mistral: Mistral Large 3 2512 — $0.5/M, score 20.1 DeepSeek V3.2 — $0.18/M, score 44.2 OpenAI: GPT-5.5 (batch) — $2.5/M, score 74.9 Anthropic: Claude Opus 4.7 — $5/M, score 73.6 Google: Gemini 3.5 Flash (batch) — $0.75/M, score 70.1 OpenAI: GPT-5.4 Nano (batch) — $0.1/M, score 56.1 Google: Gemini 2.5 Pro (batch) — $0.625/M, score 33.3 Inception: Mercury 2 — $0.25/M, score 31.1 OpenAI: GPT-4.1 Nano (batch) — $0.05/M, score 11.1 Gemini 3.6 Flash — $0.75/M, score 69.2 Anthropic: Claude Sonnet 4.5 — $3/M, score 52.1 OpenAI: GPT-4o-mini (batch) — $0.075/M, score 11.4 Qwen: Qwen3.5-9B (batch) — $0.17/M, score 28.7 GPT-5.4 mini — $0.75/M, score 56.1 Gemma 4 26B A4B IT — $0.042/M, score 39.3 Claude Fable 5 — $10/M, score 76.5 Mistral: Mistral Small 4 (batch) — $0.075/M, score 26.6 Gemini 2.5 Pro — $1.25/M, score 33.3 Z.ai: GLM 5.3 Flash — $0.15/M, score 71.5 OpenAI: GPT-3.5 Turbo (batch) — $0.25/M, score 10.7 MiniMax: MiniMax M3 — $0.3/M, score 58.6 DeepSeek: DeepSeek V4 Pro 0813 (batch) — $0.66/M, score 68.8 Thinking Machines: Inkling Small (batch) — $0.5/M, score 52.9 Anthropic: Claude Fable 5.1 — $10/M, score 81.6 Google: Gemma 4 31B (batch) — $0.39/M, score 43.4 Qwen: Qwen3.8 2.4T A95B (batch) — $2/M, score 71.9 Mistral: Devstral 2 2512 — $0.4/M, score 31.3 MoonshotAI: Kimi K2.7 Code (batch) — $0.95/M, score 60.8 DeepSeek-R1 — $0.7/M, score 24.6 Inkling — $1.87/M, score 52.1 OpenAI: gpt-oss-20b (batch) — $0.05/M, score 20.7 Muse Spark 1.1 — $1.25/M, score 71.3 MiMo-V2.5 — $0.14/M, score 56.8 Google: Gemini 3.5 Flash Lite (batch) — $0.15/M, score 49.3 Kimi K2 Thinking — $0.4/M, score 21.0 Mistral: Mistral Medium 3.5 (batch) — $0.75/M, score 46.9 Qwen: Qwen3.8 Max (0803) — $2/M, score 68.9 Mistral: Mistral Medium 3.1 (batch) — $0.2/M, score 20.5 GPT-5.1 — $1.25/M, score 49.4 Qwen: Qwen3.8 27B — $0.42/M, score 68.1 Anthropic: Claude Fable 5.1 (batch) — $5/M, score 81.6 Google: Gemini 3.8 Flash (batch) — $0.375/M, score 76.3 Google: Gemini 3.7 Flash (batch) — $0.375/M, score 76.1 inclusionAI: Ling 3.0 Flash VL — $0.06/M, score 57.0 Z.ai: GLM 5.3 (batch) — $0.7/M, score 74.8 DeepSeek: DeepSeek V4 Flash 0731 (batch) — $0.11/M, score 69.1 MoonshotAI: Kimi K3 (batch) — $3/M, score 76.2 Inkling Small — $0.45/M, score 52.9 GPT-5.5 — $5/M, score 74.9 Anthropic: Claude Sonnet 5 (batch) — $1/M, score 71.5 Anthropic: Claude Fable 5 (batch) — $5/M, score 76.5 Nemotron 3 Super 120B A12B — $0.2/M, score 37.7 IBM: Granite 4.2 8B — $0.06/M, score 22.4 Thinking Machines: Inkling (batch) — $1/M, score 52.1 GPT OSS 120B — $0.03/M, score 30.4 GPT-4 — $30/M, score 13.1 SpaceXAI: Grok 4.6 — $2/M, score 76.8 DeepSeek: DeepSeek V3.1 Terminus — $0.27/M, score 43.5 GPT-5 Mini — $0.25/M, score 15.6 Gemma 3 4B IT — $0.04/M, score 2.7 Claude Opus 5 (batch) — $2.5/M, score 78.0 OpenAI: GPT-6 Astra (batch) — $5/M, score 76.9 OpenAI: GPT-5.6 Terra (batch) — $1/M, score 76.7 OpenAI: GPT-5.4 (batch) — $1.25/M, score 71.1 DeepSeek V4 Flash — $0.15/M, score 52.0 Kimi K2.7 Code — $0.95/M, score 60.8 DeepSeek V4 Pro 0813 — $0.442/M, score 68.8 Z.ai: GLM 5.2 (batch) — $0.7/M, score 68.8 Google: Gemini 3.6 Flash (batch) — $0.375/M, score 69.2 Nemotron 3 Nano 30B A3B — $0.05/M, score 14.4 Z.ai: GLM 5.3 — $1.4/M, score 74.8 Anthropic: Claude Opus 4.8 — $5/M, score 74.3 SpaceXAI: Grok 4.3 — $1.25/M, score 42.2 GPT-5.6 Terra — $2/M, score 76.7 MiniMax: MiniMax M3 (batch) — $0.3/M, score 58.6 Gemini 3.8 Flash — $0.75/M, score 76.3 NVIDIA: Nemotron 3 Ultra (batch) — $0.6/M, score 49.3 MiniMax: MiniMax M2.7 — $0.3/M, score 52.6 SpaceXAI: Grok 4.5 — $2/M, score 72.4 Mistral: Mistral Medium 3.1 — $0.4/M, score 20.5 Kimi K2.6 — $0.95/M, score 61.8 Kimi K3 — $3/M, score 76.2 Anthropic: Claude Sonnet 4.5 (batch) — $1.5/M, score 52.1 OpenAI: gpt-oss-120b (batch) — $0.15/M, score 30.4 GPT-3.5-turbo — $0.5/M, score 10.7 Solar Pro 4 — $0.3/M, score 52.7 Gemini 3.1 Pro Preview — $2/M, score 68.8 Anthropic: Claude Opus 4.7 (batch) — $2.5/M, score 73.6 MiMo-V2.5-Pro — $0.435/M, score 60.2 Google: Gemini 3.1 Pro Preview (batch) — $1/M, score 68.8 Anthropic: Claude Sonnet 4.6 — $3/M, score 63.0 Anthropic: Claude Sonnet 4.6 (batch) — $1.5/M, score 63.0 DeepSeek V4 Flash 0731 — $0.05/M, score 69.1 GPT-5 — $1.25/M, score 37.8 OpenAI: GPT-5 Mini (batch) — $0.125/M, score 15.6 GPT-6 Astra — $10/M, score 76.9 inclusionAI: Ling 3.0 Flash — $0.021/M, score 50.6 Qwen: Qwen3.7 Plus — $0.32/M, score 55.9 SpaceXAI: Grok 4.3 (batch) — $1/M, score 42.2 GPT-4 Turbo — $10/M, score 21.5 Nemotron 3 Ultra 550B A55B — $0.5/M, score 49.3 Anthropic: Claude Sonnet 4 — $3/M, score 37.6 OpenAI: GPT-5.6 Luna (batch) — $0.1/M, score 71.4 OpenAI: GPT-5.4 Mini (batch) — $0.375/M, score 56.1 Qwen: Qwen3.8 2.4T A95B — $2/M, score 71.9 OpenAI: GPT-5.6 Sol (batch) — $1/M, score 77.4 Qwen: Qwen3.6 Plus — $0.325/M, score 54.5 Z.ai: GLM 4.6 — $0.43/M, score 45.8 GPT-4o (2024-05-13) — $5/M, score 24.2 Muse Spark 1.2 — $1.25/M, score 72.2 Qwen: Qwen3 Next 80B A3B Thinking — $0.15/M, score 17.4 GPT-5.6 Sol — $4/M, score 77.4 OpenAI: GPT-4 Turbo (batch) — $5/M, score 21.5 GPT-5.4 nano — $0.2/M, score 56.1 Nemotron 3.5 Lightning 30B A3B — $0.05/M, score 26.8 Hy3 preview — $0.066/M, score 58.8 GPT-4.1 nano — $0.1/M, score 11.1 GPT OSS 20B — $0.02/M, score 20.7 GPT-4o mini — $0.15/M, score 11.4 o1 — $15/M, score 39.7 Kimi K2.5 — $0.3/M, score 46.8 Trinity Large Thinking — $0.25/M, score 25.8 Qwen: Qwen3 30B A3B Thinking 2507 — $0.2/M, score 12.1 Z.ai: GLM 5.2 — $0.966/M, score 68.8 Qwen: Qwen3.7 Max — $1.475/M, score 66.0 SpaceXAI: Grok Build 0.1 — $1/M, score 51.5 Qwen: Qwen3.6 35B A3B — $0.1/M, score 41.9 Qwen: Qwen3.6 27B — $0.3/M, score 53.7 inclusionAI: Ling-2.6-flash — $0.01/M, score 25.3 Mistral: Mistral Small 4 — $0.15/M, score 26.6 Kwaipilot: KAT-Coder-Pro V2 — $0.3/M, score 59.5 Qwen: Qwen3.5-9B — $0.1/M, score 28.7 Qwen: Qwen3.5-35B-A3B — $0.312/M, score 37.0 Qwen: Qwen3.5-122B-A10B — $0.26/M, score 45.7 Upstage: Solar Pro 3 — $0.15/M, score 16.2 Z.ai: GLM 4.7 — $0.4/M, score 45.3 Amazon: Nova 2 Lite — $0.3/M, score 23.0 Mistral: Ministral 3 3B 2512 — $0.1/M, score 4.8 Google: Gemma 3n 4B — $0.06/M, score 3.2 Meta: Llama 4 Maverick — $0.2/M, score 16.3 OpenAI: o3 Mini High — $1.1/M, score 16.3 Meta: Llama 3.3 70B Instruct — $0.1/M, score 11.9 Meta: Llama 3.1 8B Instruct — $0.05/M, score 5.4 Nex AGI: Nex-N2-Pro — $0.25/M, score 59.1 Mistral: Mistral Medium 3.5 — $1.5/M, score 46.9 inclusionAI: Ring-2.6-1T — $0.075/M, score 42.8 IBM: Granite 4.1 8B — $0.05/M, score 9.5 Qwen: Qwen3.5 397B A17B — $0.55/M, score 48.2 Qwen: Qwen3 Coder Next — $0.12/M, score 36.2 Mistral: Ministral 3 14B 2512 — $0.2/M, score 14.4 Anthropic: Claude Haiku 4.5 — $1/M, score 43.9 Qwen: Qwen3 235B A22B Thinking 2507 — $0.23/M, score 22.1 Qwen: Qwen3 8B — $0.117/M, score 9.0 Qwen: Qwen3 14B — $0.227/M, score 13.8 Qwen: Qwen3 32B — $0.08/M, score 15.3 Meta: Llama 4 Scout — $0.1/M, score 8.2 DeepSeek: DeepSeek V3 0324 — $0.25/M, score 21.2 Cohere: Command A — $2.5/M, score 27.8 Step 3.7 Flash — $0.185/M, score 39.6 Trinity Large Thinking Input price per million tokens (log scale) Index

The stepped line is the efficient frontier: at each price, the best score available for that money or less. A model sitting on it is not being beaten by anything cheaper. Price is log-scaled because listed rates span four orders of magnitude. Only models with both a listed price and a score on this benchmark can appear.

Catalog activity

Change log

Field-level changes detected between successful source imports.

Full change log
Max Output Tokens80000 → 262144
Context Length262144 → 524288
Price Completion0.7999999999999999 → 0.9
Max Output Tokens262144 → 80000
Context Length524288 → 262144
Price Completion0.9 → 0.7999999999999999
Max Output Tokens80000 → 262144
Context Length262144 → 524288
Price Completion0.7999999999999999 → 0.9
Max Output Tokens262144 → 80000
Provenance

Sources & verification

Every figure on this page traces back to one of these records.

Methodology
Last verified
Sep 11, 2026
Status
Source-linked
Confidence
Medium
Catalog source
Models.dev
Models.devcatalog ↗Arcee documentationprovider documentation ↗Vercel AI Gateway documentationprovider documentation ↗Kilo Gateway documentationprovider documentation ↗OpenRouter documentationprovider documentation ↗NanoGPT documentationprovider documentation ↗Intelligence Indexindependent benchmark ↗Design Arena: websiteindependent benchmark ↗
Public API

Use this record

Fetch the complete source-linked model record. No key, no account, no rate-limited tier.

API documentation
Endpoint
GET https://model.kyssta.lol/api/v1/models/arcee-ai/trinity-large-thinking
curl
curl "https://model.kyssta.lol/api/v1/models/arcee-ai/trinity-large-thinking"
Common questions

Frequently asked questions

Answered directly from the stored record — nothing here is generated beyond the catalog's own fields.

What is Trinity Large Thinking?

Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use. It is published by Arcee Ai and catalogued here from Models.dev.

How much does Trinity Large Thinking cost?

Listed input pricing starts at $0.25 per million tokens from Arcee.

What is the context length of Trinity Large Thinking?

Trinity Large Thinking accepts up to 524.288K tokens of context and returns up to 262.144K output tokens.

Does Trinity Large Thinking support tool calling and structured output?

Provider catalogs list support for tool calling, and reasoning.

Which providers serve Trinity Large Thinking?

5 providers list this model: Arcee, Vercel AI Gateway, Kilo Gateway, OpenRouter, NanoGPT.

Are the weights for Trinity Large Thinking open?

Yes. The weights are published and downloadable from Hugging Face under the OpenMDW-1.1 license.

When was Trinity Large Thinking released?

The catalog records a release date of 2026-04-01, last verified Sep 11, 2026.

More models from Arcee Ai

View all →
Arcee AI: Coder Large

Coder‑Large is a 32 B‑parameter offspring of Qwen 2.5‑Instruct that has been further trained on permissively‑licensed GitHub, CodeSearchNet and synthetic bug‑fix corpora. It supports a 32k context window, enabling multi‑file...

32.768K ctx $0.5/M in Trinity Nano Preview

Experimental chat-tuned 6B MoE model with 1B active parameters for low-resource chat and instruction following

131.072K ctx Arcee AI: Virtuoso Large

Virtuoso‑Large is Arcee's top‑tier general‑purpose LLM at 72 B parameters, tuned to tackle cross‑domain reasoning, creative writing and enterprise QA. Unlike many 70 B peers, it retains the 128 k...

131.072K ctx $0.75/M in Trinity Large Preview

Lightly post-trained 398B MoE chat model for creative work, long-context prompts, and tool-using agents

524.288K ctx