Falcon-H1-34B-Instruct
- Input
- —
- Output
- —
- Speed
- ~82tok/sest.
- Context
- 262Ktokens
Hybrid attention + Mamba-2, largest Falcon-H1 size. No API price found. Falcon-H1R-7B (reasoning) is tracked separately.
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
- FP8 weights
- Throughput
- ~4,722 tok/s
- many requests batched
- Price
- $0.17–1.52 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (int4, ~75 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (33.6B params) + 25% KV-cache headroom ≈ 42 GB
- 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~4,722 output tok/s aggregate at batch 135 (bandwidth-bound)
- Blended 3:1 input:output; prefill ~31,641 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
~#17 of 21 (estimated)~64.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #16Nemotron 3 Ultra70.7%
- ~#17Falcon-H1-34B-InstructEstimated64.6%
- #17Intern-S2-Preview (35B-A3B)64.0%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~last of 24 (estimated)below measured range (<43.8%)
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #23Nex-N2.5-Mini43.8%
- ~#24Falcon-H1-34B-InstructEstimated<43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#44 of 46 (estimated)~25.3%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #43Hy328.0%
- ~#44Falcon-H1-34B-InstructEstimated25.3%
- #44Nemotron 3.5 Super VL24.6%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#53 of 102 (estimated)~0.5%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #46Qwen3.7 Plus1.0%
- ~#53Falcon-H1-34B-InstructEstimated0.5%
- #53Command A+0.5%
No published score — estimated from related benchmarks
LiveCodeBenchLab-reported
#19 of 2749.7%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #18Phi-4-reasoning-plus53.1%
- #19Falcon-H1-34B-Instruct49.7%
- #20Mistral Large 346.5%
huggingface.co · 2025-05-21
SciCodeEstimated
~#103 of 119 (estimated)~31.7%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #102Llama 4 Maverick31.7%
- ~#103Falcon-H1-34B-InstructEstimated31.7%
- #103Granite 4.2 8B31.5%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~#25 of 26 (estimated)~27.3
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #24GPT-6 Luna31.1
- ~#25Falcon-H1-34B-InstructEstimated27.3
- #25Muse Spark Open26.3
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#156 of 19249.7%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #155Llama 3.3 Instruct 70B49.8%
- #156Falcon-H1-34B-Instruct49.7%
- #157Hermes 4 - Llama-3.1 70B49.1%
huggingface.co · 2025-05-21
Humanity's Last ExamEstimated
~#131 of 199 (estimated)~7.5%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #130Sarvam 30B7.5%
- ~#131Falcon-H1-34B-InstructEstimated7.5%
- #131LFM2.5-24B-A2B7.4%
No published score — estimated from related benchmarks
AIME (latest)Lab-reported
#31 of 3316.7%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #30Llama 4 Maverick19.3%
- #31Falcon-H1-34B-Instruct16.7%
- #32Llama 4 Scout14.0%
huggingface.co · 2025-05-21
AA Intelligence IndexEstimated
~#102 of 103 (estimated)~4.4
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #100NVIDIA Nemotron 3 Nano 4B4.8
- ~#102Falcon-H1-34B-InstructEstimated4.4
- #102K2 Horizon 0.9B3.0
No published score — estimated from related benchmarks