Skip to content
Model Pareto

Xing4.0-29B-A4B

China Telecom AI (TeleAI)Open weightshuggingface.co ↗
Input
—
Output
—
Speed
~110tok/sest.
Context
262Ktokens

Released Sep 17, 2026 per the official GitHub announcement (https://github.com/XingChen-AGI/Xing4.0-29B-A4B#news). Apache-2.0 MoE, formerly the TeleChat series; native 256K context, extensible to 512K. Parameter counts follow the developer's model card. MTP supports speculative decoding; no separate official draft model was found as of Sep 26. No measured API price or speed found. Terminal-Bench 2.1 and Tau3-Bench scores are not mapped to the site's Terminal-Bench 4.0 and tau3-Banking benchmarks.

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

Hardware
1× B200 180GB
BF16 weights
Throughput
~10,211 tok/s
many requests batched
Price
$0.067–0.60 /M tok
busy → light use
On your own machine
RTX 5090 32GB (int4, 300+ tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • BF16 weights (29B params, 4B active per token) + 25% KV-cache headroom ≈ 73 GB
  • 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~10,211 output tok/s aggregate at batch 256 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~132,891 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Coding

~#16 of 24 (estimated)~58.4%
  1. #1Claude Fable 5.293.0%
  2. #2Claude Opus 5.589.9%
  3. #15GPT-5.558.6%
  4. ~#16Xing4.0-29B-A4BEstimated58.4%
  5. #16Hy357.9%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#41 of 46 (estimated)~41.7%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #40Qwen3.8 27B42.2%
  4. ~#41Xing4.0-29B-A4BEstimated41.7%
  5. #41Laguna S 2.140.4%
No published score — estimated from related benchmarks
~#24 of 102 (estimated)~29.1%
  1. #1Kimi K3.1 Pro67.2%
  2. #2Claude Sonnet 5.563.6%
  3. #22GLM-5.3 Flash32.8%
  4. ~#24Xing4.0-29B-A4BEstimated29.1%
  5. #24DeepSeek V4.1 Flash26.8%
No published score — estimated from related benchmarks
SciCodeEstimated
~#27 of 119 (estimated)~55.7%
  1. #1Claude Fable 5.267.7%
  2. #2Claude Opus 5.566.9%
  3. #26GPT-6.1 Sol55.7%
  4. ~#27Xing4.0-29B-A4BEstimated55.7%
  5. #27Claude Haiku 5.555.0%
No published score — estimated from related benchmarks
~#8 of 26 (estimated)~54.5
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #7GPT-6 Astra54.7
  4. ~#8Xing4.0-29B-A4BEstimated54.5
  5. #8GPT-6.1 Sol54.1
No published score — estimated from related benchmarks

Writing & chat

Hard reasoning