ModelsCompareBest forBenchmarksStatusPricingAPI
Compare / head-to-head

Phi-4 Multimodal InstructvsStep 3.7 Flash

Phi-4 Multimodal Instruct and Step 3.7 Flash do not yet share a public benchmark, so quality cannot be compared apples-to-apples. Step 3.7 Flash has the larger context window (256K tokens).

Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.

Shared benchmarks
0
no overlap yet
Cheaper per token
list price, input + output
Larger context
Step 3.7 Flash
256K tokens
Provider uptime (30d)
— · —
Microsoft · StepFun
Phi-4 Multimodal Instruct
Microsoft · released 2025-02-26
textvisionaudio
Context
128K
Max out
4K
Input /1M
Output /1M
Cached /1M
Scores
19 · 0 core
Step 3.7 Flash
StepFun · released 2026-05-29
textvision
Context
256K
Max out
Input /1M
Output /1M
Cached /1M
Scores
19 · 6 core
Quality

Benchmark matrix

BenchmarkPhi-4 Multimodal InstructStep 3.7 FlashΔEdge
Reported by both · 0
No public benchmark is reported for both models yet.
Only Phi-4 Multimodal Instruct reports · 19
AI2D82.3% not reported
BLINK61.3% not reported
ChartQA81.4% not reported
DocVQA93.2% not reported
InfoVQA72.7% not reported
InterGPS48.6% not reported
MathVista testmini62.4% not reported
MMBench (dev-en)86.7% not reported
MMMU55.1% not reported
MMMU-Pro (standard/vision)38.5% not reported
OCRBench84.4% not reported
POPE85.6% not reported
ScienceQA Visual97.5% not reported
Speech AI2D68.9% not reported
Speech ChartQA69% not reported
Speech DocVQA87.3% not reported
Speech InfoVQA63.7% not reported
TextVQA (val)75.6% not reported
Video-MME (16 frames)55% not reported
Only Step 3.7 Flash reports · 19
SWE-bench Verifiednot reported76.5%
AIME 2026not reported95% matharena
AA-LCRnot reported63.9%
BrowseCompnot reported75.8%
ClawEval v1.1not reported67.1%
DeepSearchQA Accuracynot reported81.7%
DeepSearchQA F1not reported92.8%
GDPval-Stirrupnot reported1415.8
GDPval-Stirrup Intelligence Indexnot reported45.8
HLE (with tools, text-only)not reported49.7%
HLE (with tools)not reported47.2%
ResearchRubricsnot reported71.7%
SimpleVQA (with tools)not reported79.2%
SWE-bench Multilingualnot reported72.4%
SWE-bench Pronot reported56.3%
Terminal-Bench 2.1not reported59.6%
Toolathlonnot reported49.5%
V* (with Python)not reported95.3%
WorldVQA (with visual search)not reported58.1%
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Phi-4 Multimodal Instruct minus Step 3.7 Flash in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing

Side by side

SpecPhi-4 Multimodal InstructStep 3.7 FlashEdge
Context window128K tokens256K tokensStep 3.7 Flash
Max output4K tokens
Input price / 1M
Output price / 1M
Cached input / 1M
Input + output / 1M
Lower is cheaper. List prices; batch, tool and regional fees excluded.
Modalitiestext · vision · audiotext · visionPhi-4 Multimodal Instruct
Released2025-02-262026-05-29
Cited benchmark scores1919
Reliability

Provider status

All providers
More matchups

Phi-4 Multimodal Instruct vs …

More matchups

Step 3.7 Flash vs …

Built by Respan
Which one wins on your data?

Public benchmarks are a starting point. Run Phi-4 Multimodal Instruct and Step 3.7 Flash on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.