Skip to content
Model Pareto

Compare models

Side by side: quality per category, price, speed, and every benchmark cell with where it came from. Pick up to four.
GLM-5.3 Flash NVFP4Step 5 PreviewTernary Bonsai 2 27B
Focus

Vision: quality vs price

Compared models are ringed; the other 87 ranked here are greyed.

Lower price is better. Pareto frontier: Gemma 4 E4B, Qwen3.5 4B, Ling-3.0-flash-VL, MiMo-V2.5, GPT-6 Luna, Qwen3.8-Omni-Flash, Gemini 3.8 Flash, GPT-6 Sol, Claude Opus 5.5, GPT-6 Astra. 23 models are not plotted: Intern-S2-397B, Qwen3.8 Flash-Next NVFP4, Ternary Bonsai 2 27B, GLM-5.3 Flash NVFP4, Qwen3.8 27B NVFP4, DeepSeek V4.1 Flash NVFP4, Intern-S2-Preview (35B-A3B), Ternary Bonsai 27B, EXAONE 4.5 33B, Bonsai 27B, DiffusionGemma 26B A4B, Step3 VL 10B, Command A+, Devstral Small 2, Qwen3.5 2B, Gemma 4 E2B, Llama 3.2 Instruct 90B (Vision), MiniCPM-V 4.6 1.3B, Molmo2-8B, LFM2.5-VL-1.6B, Qwen3.5 0.8B, Molmo 7B-D, Phi-4 Multimodal Instruct.

23 models with no price data — shown in the strip at the left edge

Intern-S2-397B, Qwen3.8 Flash-Next NVFP4, Ternary Bonsai 2 27B, GLM-5.3 Flash NVFP4, Qwen3.8 27B NVFP4, DeepSeek V4.1 Flash NVFP4, Intern-S2-Preview (35B-A3B), Ternary Bonsai 27B, EXAONE 4.5 33B, Bonsai 27B, DiffusionGemma 26B A4B, Step3 VL 10B, Command A+, Devstral Small 2, Qwen3.5 2B, Gemma 4 E2B, Llama 3.2 Instruct 90B (Vision), MiniCPM-V 4.6 1.3B, Molmo2-8B, LFM2.5-VL-1.6B, Qwen3.5 0.8B, Molmo 7B-D, Phi-4 Multimodal Instruct

  • Best-value frontier (nothing is both cheaper and better)
  • Evidencestrong → weak
  • Estimated from other categories
  • GLM-5.3 Flash NVFP4
    Quality
    68.8
    Rank
    #22/90
    Price
    —
    Speed
    —

    vs Ternary Bonsai 2 27B: −0.2 quality · price n/a

  • Step 5 Preview
    Quality
    66.1
    Rank
    #26/90
    Price
    $1.43/M tok
    Speed
    85tok/s
    In$1.00Out$2.70/M tok

    vs Ternary Bonsai 2 27B: −2.9 quality · price n/a

  • Ternary Bonsai 2 27B
    Quality
    69.0
    Rank
    #20/90
    Price
    —
    Speed
    —

    vs GLM-5.3 Flash NVFP4: +0.2 quality · price n/a

Overview

Lab
NVIDIA
StepFun
PrismML
Released
Sep 2, 2026
Sep 20, 2026
Sep 16, 2026
Weights
Open
Proprietary
Open
Input price
$/M tok
—
$1.00
—
Output price
$/M tok
—
$2.70
—
Blended price
$/M tok · 3:1 in:out
—
$1.425
—
Output speed
tok/s
—
85
—
Context
tokens
1MBest
1MBest
262K

Quality by category

0–100 within each category (100 = best tracked model). Rank is among all models ranked there. Click a row to focus it.

71.8±10.1
#12/114Lab-reported
72.7±6.4Best
#11/114Mixed
71.8±10.3
#13/114Lab-reported
67.4±9.7
#37/155Lab-reported
81.0±12.7Best
#12/155Estimated from other categories
76.9±9.7
#21/155Lab-reported
68.8±8.7
#22/90Lab-reported
66.1±6.2
#26/90Verified
69.0±8.7Best
#20/90Lab-reported
73.0±9.6
#31/193Lab-reported
83.2±4.0Best
#11/193Mixed
74.4±9.9
#28/193Lab-reported
Not ranked
75.4±6.3
#19/176Mixed
Not ranked

Vision benchmarks

Rank among models with a published score, and the leaderboard around each model. ~ italic = no published score, estimated from related benchmarks.

MMMU
day 0
~#2/3~71.3%Estimated
estimated from related benchmarks
Benchmarks in other categories (4)

Coding benchmarks

Rank among models with a published score, and the leaderboard around each model. ~ italic = no published score, estimated from related benchmarks.

Writing & chat benchmarks

Rank among models with a published score, and the leaderboard around each model. ~ italic = no published score, estimated from related benchmarks.

Hard reasoning benchmarks

Rank among models with a published score, and the leaderboard around each model. ~ italic = no published score, estimated from related benchmarks.

Agents benchmarks

Rank among models with a published score, and the leaderboard around each model. ~ italic = no published score, estimated from related benchmarks.