Intern-S2-397B
- Input
- —
- Output
- —
- Speed
- ~90tok/sest.
- Context
- 262Ktokens
Scientific multimodal MoE (397B). No API price found (not on OpenRouter). Lab-reported scores from the HF model card results table (image).
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
- 4× B200 180GB
- FP8 weights
- Throughput
- ~12,969 tok/s
- many requests batched
- Price
- $0.20–1.76 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Ultra 512GB (int4, ~99 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (397B params, 17B active per token) + 25% KV-cache headroom ≈ 496 GB
- 4× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~12,969 output tok/s aggregate at batch 371 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~250,147 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
SWE-bench VerifiedEstimated
~#4 of 21 (estimated)~81.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #3Beam 1.0 501B A23B81.6%
- ~#4Intern-S2-397BEstimated81.6%
- #4Gemini 3.1 Pro80.6%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Lab-reported
#6 of 2368.5%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #5Claude Opus 579.2%
- #6Intern-S2-397B68.5%
- #7Beam 1.0 501B A23B66.1%
huggingface.co · 2026-09-13
DeepSWE v1.1Estimated
~#37 of 46 (estimated)~51.7%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #36Nex-N2.5-Pro55.8%
- ~#37Intern-S2-397BEstimated51.7%
- #37Qwen3.8 Max (0902)51.0%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#31 of 102 (estimated)~19.9%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #30SWE-2 Flash20.0%
- ~#31Intern-S2-397BEstimated19.9%
- #31Gemini 3.8 Flash19.7%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#12 of 28 (estimated)~78.5%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #11Gemma 4 31B80.0%
- ~#12Intern-S2-397BEstimated78.5%
- #12gpt-oss-20b77.7%
No published score — estimated from related benchmarks
SciCodeLab-reported
#54 of 11847.6%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #52Nemotron 3.5 Super VL47.8%
- #54Intern-S2-397B47.6%
- #55GPT-5.4 nano47.2%
huggingface.co · 2026-09-13
AA Coding Agent IndexEstimated
~#21 of 26 (estimated)~41.0
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #20Kimi K341.4
- ~#21Intern-S2-397BEstimated41.0
- #21Qwen3.8 Max (0902)39.2
No published score — estimated from related benchmarks
Vision
MMMU-ProLab-reported
#11 of 8781.7%
- #1Claude Opus 5.587.7%
- #2GPT-6 Astra86.9%
- #10Muse Spark 1.382.0%
- #11Intern-S2-397B81.7%
- #12GPT-5.6 Terra80.7%
huggingface.co · 2026-09-13
CharXiv ReasoningEstimated
~#5 of 12 (estimated)~82.8%
- #1Gemini 3.8 Flash86.2%
- #2Kimi K384.8%
- #4Qwen3.8-Omni-Flash83.5%
- ~#5Intern-S2-397BEstimated82.8%
- #5MiniMax-M3.1-Flash82.5%
No published score — estimated from related benchmarks
OmniDocBench v1.5Estimated
~#4 of 8 (estimated)~88.4%
- #1MiniMax-M391.6%
- #2Kimi K391.1%
- #2Qwen3.8 27B91.1%
- ~#4Intern-S2-397BEstimated88.4%
- #4Gemini 3.1 Pro85.3%
No published score — estimated from related benchmarks
GDP.pdf (AA)Estimated
~#20 of 48 (estimated)~18.7%
- #1Muse Wonder Full32.6%
- #2GPT-6 Astra31.0%
- #19MiMo-V2.6-Pro19.2%
- ~#20Intern-S2-397BEstimated18.7%
- #20Gemini 3.1 Pro17.8%
No published score — estimated from related benchmarks
ChartographyEstimated
~#9 of 25 (estimated)~33.1%
- #1Claude Fable 5.273.4%
- #2GPT-6 Astra71.0%
- #8GPT-5.6 Terra34.0%
- ~#9Intern-S2-397BEstimated33.1%
- #9GPT-5.531.0%
No published score — estimated from related benchmarks