ZMIME
Comparison · 3 models · Updated Nov 2, 2026

DeepSeek R1 0528 Qwen3 8B vs Phi-4-multimodal-instruct vs Baichuan M2 32B

Phi-4-multimodal-instruct comes out ahead, 72 to 58 and 58 on our weighted score, though DeepSeek R1 0528 Qwen3 8B is 16% cheaper per token.

  1. DeepSeek

    DeepSeek R1 0528 Qwen3 8B

    Released May 29, 2025

    58/100
    • ECI—
    • Price$0.06 / $0.09
    • Context131K
    Full specs Remove DeepSeek R1 0528 Qwen3 8B
  2. Our pick

    Microsoft

    Phi-4-multimodal-instruct

    Released Feb 26, 2025

    72/100
    • ECI—
    • Price$0.07 / $0.11
    • Context131K
    Full specs Remove Phi-4-multimodal-instruct

  3. Baichuan

    Baichuan M2 32B

    Released Aug 13, 2025

    58/100
    • ECI—
    • Price$0.07 / $0.07
    • Context131K
    Full specs Remove Baichuan M2 32B
01 — Verdict

Phi-4-multimodal-instruct is our pick

Phi-4-multimodal-instruct is the better all-round choice, scoring 72/100 against DeepSeek R1 0528 Qwen3 8B (58) and Baichuan M2 32B (58). It leads on inputs & features. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

  • CapabilityNot enough dataNo independent benchmark covers every model here yet
  • Lowest priceDeepSeek R1 0528 Qwen3 8BDeepSeek R1 0528 Qwen3 8B $0.068 · Baichuan M2 32B $0.07 · Phi-4-multimodal-instruct $0.08 per 1M tokens (3:1 blend)
  • Longest contextAbout the sameDeepSeek R1 0528 Qwen3 8B 131,072 · Phi-4-multimodal-instruct 131,072 · Baichuan M2 32B 131,072 tokens
  • Widest inputsPhi-4-multimodal-instructDeepSeek R1 0528 Qwen3 8B: Text · Phi-4-multimodal-instruct: Text, Images, Audio · Baichuan M2 32B: Text
  • Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
How the score is built
MeasureWeightDeepSeek R1 0528 Qwen3 8BPhi-4-multimodal-instructBaichuan M2 32B
Price50%10098100
Inputs & features30%106010
Context window20%252525
Overall100%58/10072/10058/100

Left out because at least one model lacks the data: capability. The remaining weights were rescaled.

02 — Side by side

Every spec in one table

Highlighted cells lead their row. Dashes mean the data is not published.

DeepSeek R1 0528 Qwen3 8B vs Phi-4-multimodal-instruct vs Baichuan M2 32B specifications side by side
SpecificationDeepSeek R1 0528 Qwen3 8BDeepSeekPhi-4-multimodal-instructMicrosoftBaichuan M2 32BBaichuan
Capability
Capabilities Index (ECI)———
ECI rank———
GPQA DiamondGraduate-level science questions9.3%——
OTIS Mock AIME 2024–2025Competition mathematics43.9%——
Price per million tokens
Input$0.06 (best)$0.07$0.07
Output$0.09$0.11$0.07 (best)
Cached input———
Blended (3:1)$0.068 (best)$0.08$0.07
Long-context rateSame rateSame rateSame rate
Price sourceMedian of 1 providersMedian of 1 providersMedian of 1 providers
Limits
Context window131,072 tokens131,072 tokens131,072 tokens
Max output32,000 tokens131,072 tokens (best)131,072 tokens (best)
Inputs and features
TextYesYesYes
ImagesNoYesNo
PDFsNoNoNo
AudioNoYesNo
VideoNoNoNo
ReasoningYesNoYes
Tool callingNoYesNo
Structured outputNoNoNo
Availability
WeightsOpenMITOpenMITOpenApache-2.0
API model ID———
API providers111
ReleasedMay 29, 2025Feb 26, 2025Aug 13, 2025
Knowledge cutoff—Jun 2024—
03 — Cost

What would a month cost?

Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.

  • DeepSeek R1 0528 Qwen3 8B$0.78
  • Phi-4-multimodal-instruct$0.92
  • Baichuan M2 32B$0.84
04 — Questions

Which should you choose?

Which is better: DeepSeek R1 0528 Qwen3 8B, Phi-4-multimodal-instruct or Baichuan M2 32B?

Phi-4-multimodal-instruct is the better all-round choice, scoring 72/100 against DeepSeek R1 0528 Qwen3 8B (58) and Baichuan M2 32B (58). It leads on inputs & features. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

Which is cheaper, DeepSeek R1 0528 Qwen3 8B, Phi-4-multimodal-instruct or Baichuan M2 32B?

DeepSeek R1 0528 Qwen3 8B is cheaper at $0.06 input / $0.09 output per million tokens (median across 1 API provider). Baichuan M2 32B costs $0.07 input / $0.07 output per million tokens (median across 1 API provider); Phi-4-multimodal-instruct costs $0.07 input / $0.11 output per million tokens (median across 1 API provider). At a typical mix of three input tokens to one output token, that is $0.068 per million tokens for DeepSeek R1 0528 Qwen3 8B versus $0.07 for Baichuan M2 32B (1× as much) and $0.08 for Phi-4-multimodal-instruct (1.2× as much).

Which scores higher on benchmarks?

There is no independent benchmark that covers all three models yet. DeepSeek R1 0528 Qwen3 8B has not been scored yet, Phi-4-multimodal-instruct has not been scored yet and Baichuan M2 32B has not been scored yet.

Which is better for coding?

There are no published SWE-bench Verified results for DeepSeek R1 0528 Qwen3 8B, Phi-4-multimodal-instruct and Baichuan M2 32B yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. Note that DeepSeek R1 0528 Qwen3 8B and Baichuan M2 32B does not support tool calling, which most coding agents need.

Which has the bigger context window?

DeepSeek R1 0528 Qwen3 8B, Phi-4-multimodal-instruct and Baichuan M2 32B share the same 131,072-token context window. Maximum output per response: DeepSeek R1 0528 Qwen3 8B up to 32,000, Phi-4-multimodal-instruct up to 131,072, Baichuan M2 32B up to 131,072 tokens.

Which can read images, PDFs, audio or video?

DeepSeek R1 0528 Qwen3 8B accepts text; Phi-4-multimodal-instruct accepts text, images and audio; Baichuan M2 32B accepts text. Phi-4-multimodal-instruct handles the widest range of inputs.

Are any of these open source?

Yes, all three publish their weights (MIT and Apache-2.0), so you can self-host them.

Which is newer?

Baichuan M2 32B is the newest, released Aug 13, 2025. DeepSeek R1 0528 Qwen3 8B came out May 29, 2025; Phi-4-multimodal-instruct came out Feb 26, 2025. Knowledge cutoff: Phi-4-multimodal-instruct Jun 2024.

How do you decide the winner?

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 2, 2026.

05 — Keep comparing

Related comparisons