MiniCPM5 1B

model page · concurrency sweep 1–24 agents

MiniCPM5 1B

server model id: minicpm5-1b-claude-opus-fable5-v2-thinking
thinkingKV F16ctx 32k1B params

All-in throughput king (983 tok/s) but half its output is hidden reasoning and TTFT climbs to 52s.

Key findings

  • Reasoning dominates: this "thinking" model spends a large share of its output on hidden reasoning — ~43–52% of all tokens across runs (e.g. 30,838 of 59,018 at 24 agents). Visible content is only about half of what it actually generates.
  • All-in throughput is high: counting reasoning, combined throughput reaches 983 tok/s at 24 agents (4.3x single-agent) — well above llama-3.2-1b-mini-agent's 652 tok/s. But visible content throughput peaks much lower (~537 tok/s at 19) because of the reasoning overhead.
  • Single-agent is slow to start and short: the concurrency-1 run finished in ~24s (agent ended naturally) at 128 content / 227 all tok/s — slower than the non-thinking model's 196 tok/s.
  • TTFT is the weak point: 1.5s–52s across the sweep (mean usually 6–18s under load) — the reasoning prefill dominates first-token latency. Not suitable for latency-critical controllers at high concurrency.
  • Errors: 0 across all 24 runs.

Sweep — total concurrency 1–24

agentswall (s)contentreasontotalcomb tok/scomb allper-agentper-agent allscale %TTFT meanTTFT maxoktimeouterr
123.93,0492,3635,412128227229.6230.5100%1.5s1.5s100
260.010,65811,67322,331178372198.4222.1164%12.0s21.6s020
360.012,86911,59224,461214407175.0184.0179%11.5s22.2s120
460.017,1419,75926,900285448151.6172.2197%6.0s9.9s130
560.019,8679,61029,477331491161.0151.1216%10.3s15.6s140
660.018,67411,55630,23031150388.098.6222%17.8s47.9s060
760.018,49710,16528,66230847788.992.9210%21.9s52.2s070
860.020,53910,34330,88234251470.675.6226%14.3s15.7s080
960.017,57714,46732,04429353471.992.2235%16.8s18.1s090
1060.023,6469,13532,78139454674.485.8241%17.9s29.4s0100
1160.024,07110,81634,88740158163.871.2256%18.3s22.5s290
1260.027,07817,68744,76545174669.485.7329%5.4s26.2s660
1360.029,50513,22342,72849171268.677.5314%9.5s18.3s490
1460.027,16319,52446,68745277868.581.4343%12.3s40.0s590
1560.027,31319,16846,48145577450.061.7341%7.1s11.2s870
1660.031,41418,97250,38652383955.868.7370%5.6s10.7s880
1760.030,16719,96050,12750283554.166.9368%5.8s10.7s7100
1860.029,23721,09250,32948783852.564.9369%7.9s11.3s8100
1960.032,25919,71451,97353786648.459.0381%9.3s24.2s1270
2060.128,39526,88955,28447392153.262.4406%6.5s22.1s8120
2160.028,84724,78553,63248089348.058.1393%7.8s11.2s1290
2260.028,17828,95557,13346995251.561.0419%7.1s22.8s12100
2360.027,08428,74555,82945193048.655.1410%11.5s22.4s12110
24 peak60.028,18030,83859,01846998343.256.7433%7.0s13.2s12120

* burst artifact — agents queued ~11–16s then generated in a short burst; the per-agent timer inflates the number. Trust combined throughput and TTFT columns. Peak row = highest combined (all-token) throughput.

Charts

Whole sweep at a glance — combined throughput, scaling efficiency, per-agent throughput and outcomes across every concurrency level.
Whole sweep at a glance — combined throughput, scaling efficiency, per-agent throughput and outcomes across every concurrency level.
Combined throughput (all tokens) vs concurrent agents. Annotated peak with the concurrency level where it occurred.
Combined throughput (all tokens) vs concurrent agents. Annotated peak with the concurrency level where it occurred.
Per-agent throughput erosion as slots contend. Red stars mark burst-timing artifacts (queued ≥ 10s then generated in a burst — inflated values, not real throughput).
Per-agent throughput erosion as slots contend. Red stars mark burst-timing artifacts (queued ≥ 10s then generated in a burst — inflated values, not real throughput).
Scaling efficiency relative to the single-agent baseline (100%).
Scaling efficiency relative to the single-agent baseline (100%).
Time to first token — mean (solid) and max (dashed). Log scale. Gaps mean no visible content ever arrived within the 60s window.
Time to first token — mean (solid) and max (dashed). Log scale. Gaps mean no visible content ever arrived within the 60s window.
Total tokens per 60s run — all tokens vs visible content only.
Total tokens per 60s run — all tokens vs visible content only.
How each run of N agents finished: completed, timed out (60s budget), or errored.
How each run of N agents finished: completed, timed out (60s budget), or errored.
The concurrency trade-off on one chart — combined throughput rises while per-agent speed falls.
The concurrency trade-off on one chart — combined throughput rises while per-agent speed falls.
Visible content vs hidden reasoning tokens per run, plus the reasoning share (%).
Visible content vs hidden reasoning tokens per run, plus the reasoning share (%).

Raw data & downloads