Qwen3.5-35B-A3B qwen3.5-35b-a3b

qwen · qwen reasoning efficient

#23 overall #22 agentic

params
36B (A3B)
arch
moe
context
256k
license
apache-2.0 open weights
released
Feb 2026
reasoning
yes
downloads/30d
2.4M

# architecture

attn shared · reasoning GQA 16q/2kv router top-8 of 256 expert ×256 expert out ×40 layers ctx 262,144 36B pool · A3B/token
layers
40
d_model
2048
heads
16q / 2kv
head dim
256
experts
top-8 of 256
vocab
248320
family
qwen3_5_moe_text

signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.

# why ranked

Overall: #23 — score 64.4 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 62.9
bench_composite 0.30 68.0

benchmark panel evidence:

complete panelscore (0–100)
browsecomp-v168.0

Coding: provisional — score 42.1 ● low — a single signal; treat with caution

provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.

signalweightinput (0–100)
aa_coding 0.30 42.1
aider_polyglot 0.30
swe_bench_verified 0.40

missing signals are dropped and the remaining weights renormalized — never imputed.

Agentic: #22 — score — ● low — a single signal; treat with caution

signalweightinput (0–100)
swe_bench_verified 0.60
swe_rebench 0.40 0.0

missing signals are dropped and the remaining weights renormalized — never imputed.

full methodology

# trend

OGM score · last 54 days
OGM score over 54 days
overall rank (up is better)
Overall rank over 54 days

methodology changed during this history window; score movement across that boundary is not model movement. See methodology v5.

# benchmarks

benchmarkscoresourcedate
AA Coding Index 37.0 Artificial Analysis
AA Intelligence Index 22.6 Artificial Analysis
Aa-lcr 0.6 / 1 LLM Stats
Ai2d 0.9 / 1 LLM Stats
Androidworld-sr 0.7 / 1 LLM Stats
Babyvision 0.4 / 1 LLM Stats
Bfcl-v4 0.7 / 1 LLM Stats
Browsecomp 0.6 / 1 LLM Stats
Browsecomp-zh 0.7 / 1 LLM Stats
C-eval 0.9 / 1 LLM Stats
Cc-ocr 0.8 / 1 LLM Stats
Charxiv-r 0.8 / 1 LLM Stats
Codeforces 0.8 / 3000 LLM Stats
Countbench 1.0 LLM Stats
Vlmsareblind 1.0 / 1 LLM Stats
SWE-rebench (resolved) 17.1 SWE-rebench
SWE-rebench (pass@5) 36.9 SWE-rebench
LMArena Elo (Code) 1250.0 LMArena Aug 2026

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime
Darkbloomfp4 256k $0.08$0.75 via openrouter 99.9%
DeepInfrafp8 256k $0.14$1.00 deepinfra 99.4%
Parasailfp8 256k $0.15$1.00$0.05 via openrouter 100.0%
Alibabaunknown 256k $0.16$1.30 via openrouter 99.7%
Wandb 256k $0.25$1.25 via litellm
Veniceunknown 250k $0.31$1.25 venice 99.9%
AtlasCloudfp8 256k $0.23$1.80 atlascloud 99.8%
SiliconFlowfp8 256k $0.24$1.80 openrouter 89.8%
Novita 256k $0.25$2.00 novita 62

sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price

source aliases
llmstats
qwen3.5-35b-a3b
openrouter
qwen/qwen3.5-35b-a3b