Data as of November 17, 2026 · How the score is built
Hy4 flash
Decision readingHy4 flash is tracked, but not publicly ranked yet. The profile exposes 10 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Email me when its price, rank or availability changes.
Capability
Unranked
field median 49.4Not eligible for a public rank
Price
$0.070input / $0.28 output
input median $0.60cached $0.003 · blended $0.18
Speed
Not measured
field median 89.5 tok/sTime to first token not measured
Context
1Mtokens
field median 256,000Maximum output length is tracked separately
Decision snapshot
How much published evidence this model has, where that evidence is strongest, and what to validate before choosing it. It has no public rank, so no field position is shown.
Strongest published evidence
Published rows are visible, but no category has enough eligible evidence for a comparative rank.
Validate before choosing
10 published rows leave some tracked benchmark slots empty. Independent runtime speed has not been measured.
Source-linked · 10 displayable benchmark rows
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank Not rankedWeight 22%5 benchmarksVerified | 80.6 | 5 benchmarks | Verified | |||
| CodingWeight 20%2 benchmarksVerified | Score pending | 2 benchmarks | Verified | |||
| ReasoningWeight 17%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MultimodalRank Not rankedWeight 12%2 benchmarksVerified | 71.5 | 2 benchmarks | Verified | |||
| KnowledgeRank Not rankedWeight 12%1 benchmarkVerified | 78.2 | 1 benchmark | Verified | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MathWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured |
10 of 922 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic5/5 verified
- Coding2/2 verified
- ReasoningNot measured
- Multimodal2/2 verified
- Knowledge1/1 verified
- MultilingualNot measured
- Inst. FollowingNot measured
- MathNot measured
Benchmark ledger
Every result says who ran the test. For this model, Tencent ran all 10. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.
Coding2 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| DeepSWE | 64.6% | Best verified: Muse Wonder Full · 79.0% | Gap14.4 behind | Reported by Tencent | Shown only |
| ProgramBenchProgramBench: Can Language Models Rebuild Programs From Scratch? | 16.0% | Best verified: Claude Opus 5 · 93.0% | Gap77 behind | Reported by Tencent | Shown only |
Agentic5 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Terminal-Bench 4.0 | 19.20% | Best verified: Claude Sonnet 5.5 · 70.60% | Gap51.4 behind | Reported by Tencent | Shown only |
| AutomationBench | 52.2% | Best verified: MiniMax M3.1 Flash · 62.2% | Gap10 behind | Reported by Tencent | Shown only |
| HLE w/ toolsHumanity's Last Exam with tools | 55.2% | Best verified: Kimi K3.1 Pro · 70.0% | Gap14.8 behind | Reported by Tencent | Shown only |
| APEX-Agents | 63.1% | Best verified: Hy4 · 75.2% | Gap12.1 behind | Reported by Tencent | Shown only |
| OSWorld 2.1OSWorld 2.1 (all 108 tasks) | 68.1% | Best verified: Claude Opus 5.5 · 81.8% | Gap13.7 behind | Reported by Tencent | Shown only |
Multimodal2 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| MMMU-ProMassive Multi-discipline Multimodal Understanding Pro | 75.8% | Best verified: GPT-6.1 Terra · 85.5% | Gap9.7 behind | Reported by Tencent | Shown only |
| ZeroBench w/ PythonZeroBench_main with Python | 45.8% | Best verified: Kimi K3.1 Pro · 63.0% | Gap17.2 behind | Reported by Tencent | Shown only |
Knowledge1 row
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| HLEHumanity's Last Exam | 55.2% | Best verified: Claude Fable 5.2 · 74.2% | Gap19 behind | Reported by Tencent | Shown only |
Outlined bars are lab-reported. Filled bars are benchmark-run or independent. Bars run 0–100; the dark tick marks the best source-verified value.
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- tencent/hy4-flashTencent Cloud TokenHub model pricing
- Context window
- 1MTencent Cloud TokenHub model pricing
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- Not sourced yet
- Output modalities
- Not sourced yet
- Parameters
- Not sourced yet
- Availability
- Tencent serves Hy4 flash through Tencent Cloud TokenHub (hy4-flash) and OpenRouter (tencent/hy4-flash, served by Tencent in FP8), and in WorkBuddy, Yuanbao, WeCom, QQ, ima, and Marvis. Tencent says it plans to release Hy4 flash as an open-weight model, but no checkpoint is public yet, so source type stays pending.
- Cloud regions
- Not tracked yet
- Lifecycle
- Current
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Published at $0.003 per million cached input tokensTencent Cloud TokenHub model pricing
- Self-host
- Weight availability is pending
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track Hy4 flash, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
Hy4 flash has no verified publication yet for weight access. Its 1M context window is documented separately from that pending weight status. The profile records its reasoning mode as reasoning.
Tencent serves Hy4 flash through Tencent Cloud TokenHub (hy4-flash) and OpenRouter (tencent/hy4-flash, served by Tencent in FP8), and in WorkBuddy, Yuanbao, WeCom, QQ, ima, and Marvis. Tencent says it plans to release Hy4 flash as an open-weight model, but no checkpoint is public yet, so source type stays pending.
Official exact-value snapshot from Tencent's November 16, 2026 Hy4 flash launch post and its benchmark appendix, at the highest available reasoning setting. Tencent describes a 224B-total mixture-of-experts model, 64B of which are Engram parameters, with 12B active per token, a 1M-token context window, and native image input, distilled from Hy4 and built for cheaper everyday productivity work. We map DeepSWE v1.1, ProgramBench (almost solved), Terminal-Bench 4.0, APEX-Agents v1.1, AutomationBench v1.0.6 (600 public tasks), HLE-Full with tools, OSWorld 2.1 (full set, partial score), MMMU-Pro, and ZeroBench main with tools (pass@5) onto existing lanes. Tencent's internal WorkIndex rows (PPTGenBench, PPTCoEditBench, PPTFixBench, MdCoEditBench, WordEditBench, ExcelFixBench, AudioVideoBench, BuddyVQA Atlas), its cost-per-task columns, and the product-team evaluations quoted in the post stay out of scored fields. Tencent still recommends Hy4 for long-horizon software engineering. OpenRouter lists a 1,048,576-token context window and text plus image input.
Hy4 flash sits in the Hy4 family with Hy4, Hy4 preview. 10 of 922 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Last updated November 17, 2026. Runtime fields remain blank until a sourced snapshot exists.
Questions
How does Hy4 flash perform overall in AI benchmarks?
Hy4 flash has 10 source-displayable benchmark rows, but it does not qualify for a public overall rank. The available rows remain visible by category without being converted into a site-wide score. Missing evidence stays blank instead of being estimated from an earlier model.
Is Hy4 flash good for knowledge and understanding?
Hy4 flash has source-displayable benchmark coverage for knowledge and understanding, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Hy4 flash good for coding and programming?
Hy4 flash has source-displayable benchmark coverage for coding and programming, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Hy4 flash good for agentic tool use and computer tasks?
Hy4 flash has source-displayable benchmark coverage for agentic tool use and computer tasks, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Hy4 flash good for multimodal and grounded tasks?
Hy4 flash has source-displayable benchmark coverage for multimodal and grounded tasks, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Which sibling models are related to Hy4 flash?
Hy4 flash belongs to the Hy4 family. Related tracked variants include Hy4, Hy4 preview. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.
Does Hy4 flash have full benchmark coverage on BenchLM?
No. Hy4 flash currently has 10 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Hy4 flash?
Hy4 flash has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.