AI ScoreboardFrontend & Coding Index← Back to rankings

MODEL PROFILE / EDITION 2026.09

Tencent Hy4 preview

Cost-driven teams and self-hosters who want a permissively licensed frontier-scale coding model

TencentRank 10Confidence BNo prior snapshot
79.0

Balanced model evidence score; rank 10. No rank delta is shown because there is no prior numeric snapshot.

Best for

Cost-driven teams and self-hosters who want a permissively licensed frontier-scale coding model

Price & access

$0.834 in / $2.501 out per 1M with $0.042 cache hits (Tencent newsroom and OpenRouter agree)

Evidence summary

Strong first-party evidence (Hugging Face card with benchmark table, Tencent newsroom with exact prices) plus Arena and OpenRouter. Weakened by preview status, missing AA coverage, undocumented modalities and vendor-run benchmarks.

Edition status

No prior snapshot. Tracked in the September 2026 baseline.

Strengths

  • SWE-bench Pro 65.7% at $0.834/$2.501 is outstanding price-performance
  • Apache 2.0 with FP8 weights on four mirrors - the most permissive frontier-scale open release here
  • Arena WebDev Frontend rank 10 (1638) despite preview status
  • 1M+ context with sparse attention designed for long-horizon dev work

Tradeoffs

  • Preview-labelled: no GA commitment, and no AA coverage for independent speed/intelligence measurement
  • Modalities are not documented on the fetched pages
  • Only 1,330 Arena votes (+/-18)
  • Serving needs 8-way tensor parallelism; single-provider hosting on OpenRouter

Score profile

visual 76 · code 83 · agentic 81 · debug 78 · context 80 · speed 62 · value 93

Linked evidence

Evidence source 1 ↗Evidence source 2 ↗Evidence source 3 ↗Evidence source 4 ↗Evidence source 5 ↗
Compare this model on the full scoreboard

Use the same published data, filters, and side-by-side metrics.

Open comparison view →