Models / Qwen
Qwen3-Next-80B-A3B-Instruct
by Qwen
Used for agents, long documents and fast. Strong at calling tools.
Best forAgents and toolsLong documents
Qwen3-Next-80B-A3B-Instruct is made by Qwen. It takes text in and answers in text, with a context window of 262,144 tokens.
Price per 1M tokens
The live price, read from billing as this page loads. You pay only for the tokens you use, from prepaid credit.
catalog/qwen3-next-80b-a3b-instructTīrtha Bench review
What people use it for, in its maker's words:
What we found in our own tests:
- Agents and tool useStrong at calling tools: picks the right tool and fills in its inputs almost every time.
Our results: tool calling
Tool calling (public BFCL v4 test)
Overall: 95.6%*. Picks the right tool and fills in the inputs correctly on 96 of every 100 tasks.
One tool 98% · Picking the right tool 94% · Several calls at once 93%
A task counts as right only if the model calls the right tool with the right inputs. 800 tasks. Run on Oct 29, 2026. *Measured by Spanda Works on third-party hosts, not through Tīrtha. Results through Tīrtha may differ.
Handled several calls at once in our tests.
Try it
Model id catalog/qwen3-next-80b-a3b-instruct · base URL https://api.tirtha.ai/v1. Any OpenAI-compatible client works: change the base URL and keep your code.
Using a coding tool? Set up OpenCode, Cline, Cursor and more. PowerShell and the full reference: Quickstart.
Limits and support
- Tool callingYesAlso measured on Tīrtha Bench
- Text inYes
- StreamingYes
- Sampling settingsYou can set themThe maker suggests temperature 0.7, top p 0.8, top k 20.
What it is for
- Agents and tools
Qwen3 excels in tool calling capabilities
From the maker's card - Long documents
handling ultra-long-context tasks up to 256K tokens
From the maker's card
In the maker's own words, from its published model card. Claims and benchmarks on that card are the maker's own, not measured by Tīrtha.
What a job costs
1,000 calls of 2,000 tokens in and 500 out cost about -
Worked out from the live price above, without cached input. Your own calls will differ in size.
Similar models on Tīrtha
Licence
Weights: Apache-2.0. Read the licence.
From the maker: Model card: Qwen/Qwen3-Next-80B-A3B-Instruct on Hugging Face · Last updated by the maker: Sep 17, 2025Facts from the maker's card are the maker's own, not measured by Tīrtha.