How does Hy3 rank among AI models?
Hy3 is a superseded version and is not on the current board as of 6 November 2026. Its archive position, where one exists, is shown on the page; the newer version carries the current rank.
Tencent·#75 of 504 in the all-versions archive
Hy3 is a language model from Tencent. It ranks #75 of 504 in ModelCap's all-versions archive, with an Index score of 72.1, measured on 3 public benchmarks. Its score range of 67.3–76.9 spans positions #46–#103. It is listed at $0.132 per million input tokens and $0.528 per million output tokens across 5 providers. It reads up to 262K tokens of context, and its weights are openly downloadable.
Figures as of 6 Nov 2026, 13:52 UTC · download the public dataset · methodology
ModelCap Index · All-versions archive
72.1#75
Measured on 3 public benchmarks · range 67.3–76.9
Evidence as of 4 Nov 2026, 00:00 UTC
Superseded by Hy4 preview. It no longer holds a current-board position; its all-versions archive rank is #75.
Input
$0.132
per 1M tokens
Output
$0.528
per 1M tokens
Context
262K
tokens
Providers
5
endpoints
Providers
Uptime measured over the last 30 minutesHugging Face Inference Providers
Exact repository identityProvider economics and telemetry published by the Hugging Face Router. These are operational signals, not benchmark results.
Reasoning efforts
Best supported result can inform ModelCapPublished Max, X-High, High and other configurations for this model. These are sourced scores, not head-to-head essays.
Reported evaluation leads (11)
Not independent measurementStructured rows reported in this exact model repository at the pinned revision below. They are the publisher talking about itself: never measured evidence, never a score input on their own. They remain quarantined until dataset revision, harness, configuration and identity pass ModelCap’s independent admission review.
Other versions (3)
ModelCap Index · All-versions archive
Methodology72.1
Measured on 3 public benchmarks
All-versions archive
#75 of 504
- Score range
- 67.3–76.9
- Rank range
- #46–#103
- Evidence as of
- 4 Nov 2026, 00:00 UTC
- Benchmarks used
- Arena Coding, Arena Overall, LMArena Agent
Technical details
- Public rank basis
- Measured (measured)
- Evidence label
- Measured · 3 public benchmark observations across 3 boards
- Index support
- 65.4% · strong
- Identity binding
- Exact catalogue product · tencent/tencent/hy3 · aggregates disclosed benchmark configurations · endpoint configuration not claimed · artifact metadata tencent/Hy3@a960ebc3da325ba167f069f76c41eb62c9280d22 (not the evaluation revision)
- Index observations used
- 3
- Qualified Index evidence point
- 72.1
Bayesian Evidence Score · measured evidence input only
BES normalizes admitted public benchmark evidence for measured Index rows. Its legacy score and observables prior do not define the public language rank.
- Observed capability
- 72.1
- Evidence status
- confirmed · 88% mass
- General preference
- 73.7
- Coding
- 70.5
- Agents & tools
- 25.7
- Reasoning
- —
- Evidence breadth
- 97%
- Evidence coverage
- 65%
Market and catalogue signals · secondary, never the language Index rank
- Usage (OpenRouter popularity)
- 52.4
- Liquidity (providers × uptime)
- 77.1
- Open reach (HF downloads)
- 73.6
- Surface (context / tools / modalities)
- 75.3
- Freshness
- 62.6
- · OpenRouter popularity #16
- · 5 live providers
- · HF 30d downloads 1214836
- · Catalogue: context 262144, reasoning, tools
- · First seen 2026-07-06
These adoption and deployment observations remain context only. They do not change this model's ModelCap Index score or rank.
BES evidence selected · measured input only
- Arena Overall73.7
- Arena Coding70.5
- LMArena Agent25.7
The bounded 0–100 capability score blends each source’s competitive placement with its published achievement, then combines capability families. Evidence breadth adds a modest uncertainty adjustment; price and popularity are not part of benchmark-led rank.
Deployment readiness
Metadata index90.1
strong · 94% metadata coverage
- Artifact reproducibility
- 85.0
- Access & legal clarity
- 100.0
- Deployability
- 97.9
- Evaluation provenance
- 76.1
Not yet published: the base model it derives from, pinned evaluation dataset versions.
A metadata completeness and deployability index—not a safety certification, quality grade, or production-readiness claim.
Market activity
Venue coverage- Market Gravity
- 62.3
- OpenRouter weekly popularity
- #16 of 339
Market Gravity uses OpenRouter’s full-catalogue popularity order. Sparse Vercel top-ten observations appear only when published and do not affect the score.
Arena preference
- Status
- Ranked
- Source
- LMArena
- Official rank
- #75 of 419
- Rating
- 1454.5
- 95% confidence interval
- 1449–1460
- Votes
- 14K
- Source published at
- 28 Oct 2026, 00:00 UTC
- Category
- overall
- Identity match confidence
- 100%
Market Gravity breakdown
- Usage55%
- 52.4
- Liquidity25%
- 77.1
- Open reach15%
- 73.6
- Freshness5%
- 62.6
OpenRouter weekly popularity across the full model catalogue.
Independent providers versus the model's open or closed cohort, weighted by uptime.
Hugging Face 30-day downloads for open models; neutral for closed models.
Time since first public availability, on a six-month half-life.
Specification
- Model ID
tencent/hy3- Context window
- 262K tokens
- Declared output cap
- 128K tokens
OpenRouter catalogue or provider-endpoint declaration. Limits can vary by provider and prompt length; a model-wide maximum has not been independently verified. OpenRouter metadata
- Inputs
- Text
- Outputs
- Text
- Tokenizer
- Other
- Cached input tokens
- $0.033 / 1M
- First seen on OpenRouter
- 6 Jul 2026
Capabilities
- Supported: Reasoning
- Supported: Tool use
- Supported: Structured output
- Supported: Response format
- Not supported: Moderated
Weights & access
Open weights. Downloadable weights are published under an OSI or free-culture license in the current repository metadata. This describes weight availability, not whether the full training stack qualifies as Open Source AI.
- Access
- Open weights
- Licence
- apache-2.0
- Pinned revision
- a960ebc3da32
- Parameters
- 298.8B
- Architecture
- HYV3ForCausalLM
- Model type
- hy_v3
- Hugging Face downloads (30d)
- 1.2M
- Downloads (all time)
- 1.7M
- Likes
- 990
- Repository
- tencent/Hy3
- Repository updated
- 20 Jul 2026, 07:01 UTC
Benchmark leaderboards featuring Hy3
All leaderboardsHy3: common questions
How much does Hy3 cost per 1M tokens?
Hy3 is listed at $0.132 per 1M input tokens and $0.528 per 1M output tokens as of 6 November 2026 across 5 providers. Prices come from the OpenRouter catalogue and re-observe on every refresh. For 1,000 requests using 2,000 input and 500 output tokens each, the listed rates imply $0.528 total (2M input + 0.5M output). This excludes caching, batch discounts, prompt-length tiers, tool charges and retries; verify the chosen endpoint's rate and limits before budgeting.
What is the context window of Hy3?
Hy3 has a published context window of 262K tokens (262,144). The snapshot lists a 128,000 token catalogue-declared route output cap. OpenRouter catalogue or provider-endpoint declaration. Limits can vary by provider and prompt length; a model-wide maximum has not been independently verified. The providers table lists each endpoint's context limit.
Is Hy3 open-weight?
Open weights (apache-2.0). Downloadable weights are published under an OSI or free-culture license in the current repository metadata. This describes weight availability, not whether the full training stack qualifies as Open Source AI. The repository is tencent/Hy3 on Hugging Face.
Which API providers serve Hy3?
5 providers serve Hy3 through OpenRouter as of 6 November 2026: Tencent, Novita, GMICloud, Phala and AtlasCloud. Each provider's price, context limit, quantization and measured uptime are in the providers table above.
Which benchmarks has Hy3 been evaluated on?
Hy3 has published results on 4 public boards tracked by ModelCap as of 6 November 2026: Arena text 1454 (#75 of 419), Arena coding 1503 (#73 of 414), LMArena Agent -7.7% (#49 of 58) and WildClawBench OpenClaw 49.7 (#13 of 34). Each board page ranks every tracked model on that benchmark. The ModelCap Index scores the results its methodology admits, with published uncertainty, and shows reference boards without scoring them.
When was Hy3 released?
Hy3 first appeared in the catalogue on 6 Jul 2026. It has since been superseded by Hy4 preview. Rank and price movements since then are recorded on the site's changes feed.