Models / Hy3

Tencent

Hy3

Hy3 is strongest at writing; worth a look for simple, high-volume work where cost matters most. The cheapest model of similar intelligence is DeepSeek-V4-Flash (0423), at $0.15 per million tokens against $0.25.

Superseded. Hy4 Preview is Tencent's newer model in this line, at $1.25 blended: see Hy4 Preview.

Results 16 Nov 2026 · 22 results · 2 sources

Test Hy3 on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Price per million tokens
$0.14 in · $0.58 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices
Speed
83 tokens a second; on its test prompt the answer starts after 25.8 s, thinking time included
Artificial Analysis, on its usual API
Intelligence Index
25.3
Artificial Analysis
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
Tencent, based in China
Weights published
Released
6 Jul 2026 (Artificial Analysis)
262,144 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 72nd of 185Claude Opus 5.5 – $0.25 blended, 23rd percentilecheapest: Mistral Nemo 41st of 65Nemotron 3.5 Super VL
Models priced $0.25–1 per million tokens 13th of 61MiMo-V2.6-Pro – sets this group 16th of 20Gemini 3.5 Flash-Lite
Models with similar intelligence sets this group – $0.25 blended, 5th percentilecheapest: DeepSeek-V4-Flash (0423) 7th of 10Nemotron 3.5 Super VL
Models writing 50–100 tokens a second 11th of 21Claude Opus 5.5 – $0.25 blended, 30th percentilecheapest: Granite 4.2 8B sets this group
Models from developers based in China 26th of 75MiMo-V2.6-Pro – $0.25 blended, 18th percentilecheapest: Qwen3.5-9B 12th of 23DeepSeek-V4.1-Flash
Open-weights models 27th of 104MiMo-V2.6-Pro – $0.25 blended, 33rd percentilecheapest: Mistral Nemo 25th of 43Nemotron 3.5 Super VL

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured, then those priced $0.25–1 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live Not measured
  • CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is Not measured
  • DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled Not measured
  • SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending Not measured
  • ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 70th of 159 available · 10th of 47 at its price
BenchmarkHy3Rank among availableBest available
Artificial AnalysisArtificial Analysis: Terminal-Bench 4.0 · reported 0.5% 62nd of 107 Kimi K3.1 Pro 67.2%
Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported 64.4% 41st of 107 Claude Fable 5.1 91.4%
Artificial AnalysisArtificial Analysis: τ-bench banking · reported 22.9% 41st of 104 Grok 4.6 50.7%
  • tau2-bench: pending. Its latest results are dated 26 May 2026, before this model was released.
  • Berkeley Function Calling Leaderboard (BFCL) V4: pending. Its latest results are dated 16 Dec 2025, before this model was released.
  • OpenHands Index: pending. Its latest results are dated 30 Jun 2026, before this model was released.
  • Microsoft STATE-Bench: pending. Its latest results are dated 29 May 2026, before this model was released.
  • Vending-Bench 2: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge 67th of 187 available · 10th of 61 at its price
BenchmarkHy3Rank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 25.3 72nd of 185 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 33.5% 69th of 186 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 89.7% 43rd of 168 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 46th of 169 available · 4th of 58 at its price
BenchmarkHy3Rank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,456 44th of 153 Gemini 4 Argon 1,525
  • UGI Leaderboard: has not measured this model.
Sticking to the factsSummarising without inventing things, and factual answers 69th of 139 available · 13th of 40 at its price
BenchmarkHy3Rank among availableBest available
Arena (formerly LMArena)Arena Text factuality: overall · reported 1,445 56th of 109 Gemini 4 Argon 1,511
  • Vectara Hallucination Leaderboard: has not measured this model.
  • SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Nearest alternatives

Against Hy3 at $0.25 blended, Intelligence Index 25.3. "Similar intelligence" means within 4 points or better.

Where to run it

5 hosts on OpenRouter at the standard service tier, with a median of $0.25 blended per million tokens; quantisation where the host states it: FP8, BF16. Cheapest first.

HostQuantisationInputOutputBlended
TencentFP8$0.13$0.53$0.23
GMICloudBF16$0.14$0.58$0.25
Novitanot stated$0.14$0.58$0.25
Phalanot stated$0.15$0.64$0.27
AtlasCloudFP8$0.20$0.80$0.35

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

Hy3MiMo-V2.6-Pro
most intelligent at $0.25–1 per million tokens
DeepSeek-V4-Flash (0423)
cheapest with similar intelligence
Nemotron 3.5 Super VL
fastest with similar intelligence
Claude Opus 5.5
most intelligent available today
Price per million tokens $0.25 $0.54$0.15$0.24$8.00
Tokens a second 83 43–41196
Intelligence Index 25.3 46.324.426.557.6
Spring Prompt overall – –––82
Rank among available models, by use case
Product listings – –––13th
Decks from an analysis – –––7th
User surveys – –––4th
Marketing planning – –––9th
Agents and tool use 70th of 159 23rd59th132nd3rd
Professional work – –––3rd
Reasoning and knowledge 67th of 187 13th69th74th1st
Writing 46th of 169 23rd63rd–5th
Sticking to the facts 69th of 139 10th81st–3rd
Speed under pressure – ––––

Shaded figures are better than Hy3's. A dash means no result.

Where it has been measured

3rd cheapest of the 40 models with similar intelligence. 13th most intelligent of the 61 models priced $0.25–1 per million tokens. Among models available today, mid-field on every use case measured so far.

Every result

22 published results from 2 sources, each in the source's own units, with the configuration that produced it.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 11 results
MeasureValueRankConfigurationDated
Arena Agent: confirmed task successIPS effect estimate, higher is better -0.11
-0.14–-0.08
45 of 50 Hy3 (arena agent) 2 Oct 2026
Arena Agent: praise over complaintIPS effect estimate, higher is better -0.07
-0.10–-0.03
39 of 50 Hy3 (arena agent) 2 Oct 2026
Arena Agent: steerabilityIPS effect estimate, higher is better 0.01
-0.01–0.02
30 of 50 Hy3 (arena agent) 2 Oct 2026
Arena Agent: tool groundingIPS effect estimate, higher is better -0.03
-0.04–-0.02
49 of 50 Hy3 (arena agent) 2 Oct 2026
Arena Text factuality: overallArena rating, higher is better 1,445
1,440–1,451
78 of 182 Hy3 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,448
1,434–1,461
80 of 406 Hy3 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,422
1,408–1,435
81 of 411 Hy3 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,488
1,471–1,506
62 of 363 Hy3 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,446
1,436–1,456
72 of 413 Hy3 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,456
1,449–1,463
67 of 413 Hy3 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,427
1,416–1,439
83 of 412 Hy3 2 Oct 2026

Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Artificial Analysis · reported by Artificial Analysis · 11 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 58.8 76 of 187 Hy3 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 25.3 151 of 413 Hy3 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 89.7% 71 of 360 Hy3 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 33.5% 132 of 419 Hy3 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 79.0% 102 of 407 Hy3 20 Oct 2026
Artificial Analysis: Output speedtokens per second, higher is better 83.3 80 of 137 Hy3 20 Oct 2026
Artificial Analysis: SciCode% of problems, higher is better 48.6% 125 of 205 Hy3 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 64.4% 80 of 188 Hy3 20 Oct 2026
Artificial Analysis: Terminal-Bench 4.0% of tasks, higher is better 0.5% 143 of 204 Hy3 20 Oct 2026
Artificial Analysis: Time to first answer tokenseconds, lower is better 25.8 91 of 137 Hy3 20 Oct 2026
Artificial Analysis: τ-bench banking% of tasks, higher is better 22.9% 81 of 175 Hy3 20 Oct 2026

Not shown: Not your codebase or tools.

Will Hy3 work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.