Amber 4B (MLukauskas)v1.7 · 10 s rule
Data: JevBench v1.8.2, published 2026-10-30 · method v1.7 (10 s rule), measured under the rule
Capability 56.1 · Not ranked in the current release
Measured on the 27 Oct 2026 draw, which is not yet equated to the reference scale; stronger anchor runs for that draw are being measured. Shown unranked until then.
Last measured: 2026-10-30 · measurement release: v1.8.2
CC-BY-NC-4.0 (model-card declaration; no LICENSE file)
intelligence
27.6
calibration
84.6
speed
93.4
cost
86.4
Composite (secondary)
17.4 · Not ranked on this board
Cost and measurement conditions
$0.0028 per 1,000 decisions (estimated)
GPU-time estimate from v1.8.0: one monthly-rented RTX PRO 6000 96 GB, EUR 999/month / 730 hours, ECB USD 1.1133/EUR dated 28 Oct 2026, USD 1.5235/hour; average load 70%. Architecture: small qwen, 4B active parameters; throughput 61525 tokens/s (architecture-class calibration, 29 Oct 2026, efficient-engine prefill; capped at 100,000 tokens/s). The row's own scored cost basis: 429,428 input + 0 output tokens over 1,489 decisions. Generated tokens are counted at the same throughput as prefill. GPU-time cost USD 0.00283390936235731/1,000; old list-price estimate USD 0.00865200805910007/1,000 (deepinfra:Qwen/Qwen3.5-4B at USD 0.03/M input and USD 0/M output). Applied min(old list-price estimate, GPU-time cost) = USD 0.00283390936235731/1,000. Estimated, not a bill. Reference deepinfra:Qwen/Qwen3.5-4B is frozen at the 25 Sep cut-off and was deprecated afterwards (replaced by Qwen/Qwen3.5-9B); re-score if it changes (I-3). USD 0.00283390936 per 1,000 decisions on the frozen 1,489-item common set of this draw (USD 0.00541156958 over all 1,500 items).
p50 latency: 0.2 s
Self-hosted latency estimate under the shared method (assumption, not measured)
Measured 30 Oct 2026 on an evaluator-owned Verda VM in FIN-03 (Finland, EU), one NVIDIA RTX PRO 6000 Blackwell Server Edition, 97,887 MiB reported, driver 580.178.04, image 24.04.cuda13.0. The unchanged author decoder at 71c37554758f6df3034aac5c632030a1216a7392 ran through a first-party loopback HTTP shim with local hash-verified base weights and adapter, read-only mounts, empty credentials, offline flags and network cut before sealed input. Native rendering and tokenization were unchanged. Torch 2.14.0+cu130, transformers 5.17.0, tokenizers 0.23.3 and FLA 0.5.2 were pinned; the offline counter used tokenizers 0.23.2, with bounded public usage equality rather than a universal stack-equivalence claim. Eleven model files and runtime/source/tokenizer hashes were verified. Fresh timed public noul probes covered 256 through 82,000 tokens after separate warm-ups; the longest 82,000-token probe took 8.161467097000013 s. Choice, 255-choice and Unicode-JSON probes passed with usage equality; unsupported score returned HTTP 422. These timings are synthetic, not scored percentiles. The complete run took 79.13818168640137 s; raw output sha256 04fa8a232285878b9f5b357232fb55c8dfb3fd300449cdede8a21aebc8810972 was independently verified. Replay and ready health receipts were retained, while a separate stop-diagnostic file was missing. Provider deletion, reservation release and shredding of temporary node and local copies were confirmed. GPU rental spend is separate from the benchmark's model-cost estimate.
Measured latency over 1,500 decisions: p50 0.026 s · p95 0.037 s
1 decision over 10 s (1 of them answered without error, counted as wrong).
A decision that takes longer than 10 seconds (measured time) counts as a wrong answer and as zero calibration for that item: each type's calibration is multiplied by the share of its decisions answered within 10 s. The limit uses the measured time (raw). The displayed adjusted time for self-hosted rows (2 × + 0.15 s) is an assumption and is not used for the limit.