ZMIME
Comparison · 3 models · Updated Nov 18, 2026

Qwen2.5-VL 72B Instruct vs Claude Haiku 3 vs Aion-RP 1.0 (8B)

No independent capability benchmark covers every model here yet. The score uses price, inputs & features and context window.

Price in rating: Included · Exclude price

  1. Alibaba (Qwen)

    Qwen2.5-VL 72B Instruct

    Released Jan 26, 2025

    45/100
    • ECI—
    • Price$1.01 / $1.01
    • Context131K
  2. Anthropic

    Claude Haiku 3

    Released Mar 13, 2024

    57/100
    • ECI117.8
    • Price$0.25 / $1.25
    • Context200K
  3. Aion Labs

    Aion-RP 1.0 (8B)

    Released Feb 4, 2025

    26/100
    • ECI—
    • Price$0.80 / $1.60
    • Context33K
01 — Verdict

Capability evidence is incomplete

There is not enough shared capability evidence to name an overall winner. The scores (Claude Haiku 3 57/100, Qwen2.5-VL 72B Instruct 45/100, Aion-RP 1.0 (8B) 26/100) compare the available specifications and price. 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 priceClaude Haiku 3Claude Haiku 3 $0.50 · Aion-RP 1.0 (8B) $1.00 · Qwen2.5-VL 72B Instruct $1.01 per 1M tokens (3:1 blend)
  • Longest contextClaude Haiku 3Claude Haiku 3 200,000 · Qwen2.5-VL 72B Instruct 131,072 · Aion-RP 1.0 (8B) 32,768 tokens
  • Widest inputsClaude Haiku 3Qwen2.5-VL 72B Instruct: Text, Images · Claude Haiku 3: Text, Images, PDFs · Aion-RP 1.0 (8B): Text
  • Self-hostingQwen2.5-VL 72B Instruct and Aion-RP 1.0 (8B)Publishes downloadable weights
How the score is built
MeasureWeightQwen2.5-VL 72B InstructClaude Haiku 3Aion-RP 1.0 (8B)
Price50%486148
Inputs & features30%50600
Context window20%323911
Specs & price only100%45/10057/10026/100

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.

Qwen2.5-VL 72B Instruct vs Claude Haiku 3 vs Aion-RP 1.0 (8B) specifications side by side
SpecificationQwen2.5-VL 72B InstructAlibaba (Qwen)Claude Haiku 3AnthropicAion-RP 1.0 (8B)Aion Labs
Capability
Capabilities Index (ECI)—117.8—
ECI rank—#154 of 158—
GPQA DiamondGraduate-level science questions—36.3%—
OTIS Mock AIME 2024–2025Competition mathematics—1.8%—
Price per million tokens
Input$1.01$0.25 (best)$0.80
Output$1.01 (best)$1.25$1.60
Cached input———
Blended (3:1)$1.01$0.50 (best)$1.00
Long-context rateSame rateSame rateSame rate
Price sourceMedian of 5 providersMedian of 3 providersMedian of 4 providers
Limits
Context window131,072 tokens200,000 tokens (best)32,768 tokens
Max output8,192 tokens4,096 tokens29,491 tokens (best)
Inputs and features
TextYesYesYes
ImagesYesYesNo
PDFsNoYesNo
AudioNoNoNo
VideoNoNoNo
ReasoningNoNoNo
Tool callingYesYesNo
Structured outputNoNoNo
Availability
WeightsOpenProprietaryOpen
API model ID———
API providers5 (best)34
ReleasedJan 26, 2025Mar 13, 2024Feb 4, 2025
Knowledge cutoffApr 2024Aug 31, 2023Dec 31, 2023
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.

  • Qwen2.5-VL 72B Instruct$12.12
  • Claude Haiku 3$5.00
  • Aion-RP 1.0 (8B)$11.20
04 — Questions

Which should you choose?

Which is better: Qwen2.5-VL 72B Instruct, Claude Haiku 3 or Aion-RP 1.0 (8B)?

There is not enough shared capability evidence to name an overall winner. The scores (Claude Haiku 3 57/100, Qwen2.5-VL 72B Instruct 45/100, Aion-RP 1.0 (8B) 26/100) compare the available specifications and price. 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, Qwen2.5-VL 72B Instruct, Claude Haiku 3 or Aion-RP 1.0 (8B)?

Claude Haiku 3 is cheaper at $0.25 input / $1.25 output per million tokens (median across 3 API providers). Aion-RP 1.0 (8B) costs $0.80 input / $1.60 output per million tokens (median across 4 API providers); Qwen2.5-VL 72B Instruct costs $1.01 input / $1.01 output per million tokens (median across 5 API providers). At a typical mix of three input tokens to one output token, that is $0.50 per million tokens for Claude Haiku 3 versus $1.00 for Aion-RP 1.0 (8B) (2× as much) and $1.01 for Qwen2.5-VL 72B Instruct (2× as much).

Which scores higher on benchmarks?

There is no independent benchmark that covers all three models yet. Qwen2.5-VL 72B Instruct has not been scored yet, Claude Haiku 3 has an ECI of 117.8 and Aion-RP 1.0 (8B) has not been scored yet.

Which is better for coding?

There are no published SWE-bench Verified results for Qwen2.5-VL 72B Instruct, Claude Haiku 3 and Aion-RP 1.0 (8B) yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. Note that Aion-RP 1.0 (8B) does not support tool calling, which most coding agents need.

Which has the bigger context window?

Claude Haiku 3 has the largest context window at 200,000 tokens, against 131,072 for Qwen2.5-VL 72B Instruct and 32,768 for Aion-RP 1.0 (8B). Maximum output per response: Qwen2.5-VL 72B Instruct up to 8,192, Claude Haiku 3 up to 4,096, Aion-RP 1.0 (8B) up to 29,491 tokens.

Which can read images, PDFs, audio or video?

Qwen2.5-VL 72B Instruct accepts text and images; Claude Haiku 3 accepts text, images and PDFs; Aion-RP 1.0 (8B) accepts text. Claude Haiku 3 handles the widest range of inputs.

Are any of these open source?

Qwen2.5-VL 72B Instruct and Aion-RP 1.0 (8B) publishes its weights and can be self-hosted; Claude Haiku 3 is proprietary.

Which is newer?

Aion-RP 1.0 (8B) is the newest, released Feb 4, 2025. Qwen2.5-VL 72B Instruct came out Jan 26, 2025; Claude Haiku 3 came out Mar 13, 2024. Knowledge cutoff: Qwen2.5-VL 72B Instruct Apr 2024, Claude Haiku 3 Aug 31, 2023, Aion-RP 1.0 (8B) Dec 31, 2023.

How do you decide the winner?

This rating uses price (50%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (30%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (20%, 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 18, 2026.