model · live

Skywork-Reward-V2-Qwen3-0.6B

A series of eight reward models designed for versatility across a wide range of tasks

Skywork’s own description of the model, from Skywork-Reward-V2-Qwen3-0.6B model card (pinned revision). Served here as skywork/skywork-reward-v2-qwen3-0.6b.

On this roster: Pick it to rank: score candidate answers and keep the best, at a flat price per record instead of per token.

Skywork

What Skywork built it for

in the vendor’s words

Quoted from Skywork-Reward-V2-Qwen3-0.6B model card (pinned revision), retrieved 18 Sept 2026.

Published benchmarks

Skywork’s figures, not ours

Skywork-Reward-V2-Qwen3-0.6B: scores as published by Skywork, retrieved 18 Sept 2026
Benchmark	What it measures	Score	How Skywork ran it
RewardBench v2	Preference scoring	61.3	Vendor model-card row for Skywork-Reward-V2-Qwen3-0.6B; not a measurement of this hosted endpoint..

Every score above is Skywork’s own published result for this model, copied from Skywork-Reward-V2-Qwen3-0.6B model card (pinned revision) on 18 Sept 2026. We did not run these benchmarks and we do not restate them as our own. Each model here shows only its own maker’s table, for that reason.

Vendor usage notes

what Skywork recommends

Skywork has not published usage guidance beyond the card above.

Quoted from Skywork-Reward-V2-Qwen3-0.6B model card (pinned revision), retrieved 18 Sept 2026.

Capacity on this endpoint

dedicated capacity, not a speed claim

context8Ktokens, the served window

concurrency8in flight, before a retryable 429

No speed figure is claimed here: this surface returns a completed specialist result, not a token stream. This surface has dedicated capacity: it is admitted up to the advertised concurrency without yielding to the generation lanes, and only beyond that concurrency does the lane cap shed with a retryable 429. That posture is the price basis.

Price

per 1,000 records · no request fee · no subscription

Rates: Skywork-Reward-V2-Qwen3-0.6B
per 1,000 records
$5.00

Each item bills one record per 1,000 characters of rendered text, including chat-template characters, rounded up per item. A batch bills the sum; no minimum or per-request fee. Full price sheet and worked examples.

Capabilities

skywork/skywork-reward-v2-qwen3-0.6b

One endpoint, one base URL

POST https://api.tiyuvta.ai/v1/classify. Point an existingJSON HTTP client at it and set the model id; there is no product-specific SDK.

Conversations in, judge scores out; higher is better

text -> score. There is no completion token and no streaming: the response is data[] with an index and unbounded score per item, higher is better, not a probability, returned once the request finishes.

No tools, no structured output, no reasoning channel

Those are chat-model concepts; this surface takes text items in and scores out, nothing else.

No prompt caching

Each item bills one record per 1,000 characters of rendered text, including chat-template characters, rounded up per item. A batch bills the sum; no minimum or per-request fee.

Gated before it serves

Every build of this model is reconciled against the vendor’s published model and passes a live readiness and authentication gate before it takes traffic; a failing gate blocks the release.

Where it fits best

use it for

Best-of-N and response ranking in agent loops: score N candidates, keep the top one
Judge scoring of prompt-response pairs against a general preference model (a Bradley-Terry scalar, higher is better)
Batch evaluation: each item rounds up to whole records on rendered text, including chat-template characters; a batch bills the sum, no minimum or per-request fee

Also on the same key and balance: GLM-5.3-Flash, DeepSeek-V4.1-Flash, Qwen3-Embedding-8B (16K served context), Qwen3-Reranker-8B (16K served context), Qwen3Guard-Gen-0.6B (8K served context).

First request

skywork/skywork-reward-v2-qwen3-0.6b

shell · defaulthttps://api.tiyuvta.ai/v1

curl https://api.tiyuvta.ai/v1/classify \
  -H "Authorization: Bearer $TIYUVTA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"skywork/skywork-reward-v2-qwen3-0.6b","input":[{"messages":[{"role":"user","content":"What is the capital of France?"},{"role":"assistant","content":"Paris."}]}]}'

The request above hits /v1/classify directly; there is no sampling parameter on this surface.

Get an API key Agent client configurations Read the docs

Vendor sources: huggingface.co/Skywork/Skywork-Reward-V2-Qwen3-0.6B, Apache-2.0. Skywork trained this model; we serve it at 8K of the judge window, below its 16K native context, on dedicated specialist capacity. Benchmark scores and usage recommendations on this page come from Skywork-Reward-V2-Qwen3-0.6B model card (pinned revision), retrieved 18 Sept 2026.
Machine-readable metadata: GET https://api.tiyuvta.ai/v1/models for supported parameters, and the provider feed at GET https://api.tiyuvta.ai/models?schema=openrouter.
Rate limits: This surface has dedicated capacity, admitted up to 8 concurrent requests without yielding to the generation lanes; beyond that it sheds with a retryable 429 rather than sitting behind a per-account token ceiling.