Skip to content
OpenKey

Gemini 2.5 Flash vs Qwen3 235B A22B Instruct 2507

GoogleQwenboth via one key, provider price + 3%

Both models shipped within weeks of each other in mid-2025, but they're built for different jobs. Gemini 2.5 Flash is Google's reasoning-and-multimodal workhorse with a 1,048,576-token context window. Qwen3 235B A22B Instruct 2507 is a 235B-parameter mixture-of-experts model (22B active) that's text-only but priced for scale. Both run on OpenKey with one key and a flat 3% fee on top of provider list price — no separate accounts needed.

Spec vs spec

SpecGemini 2.5 FlashQwen3 235B A22B Instruct 2507
Context window1.0M262K
Max output66K16K
Input modalitiesfile, image, text, audio, videotext
Output modalitiestexttext
Knowledge cutoffJan 31, 2025Jun 30, 2025
ReleasedJun 17, 2025Jul 21, 2025
Reasoningoptional

Pricing

Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.

openkey.ai

google/gemini-2.5-flash

Input · 1M tokens

$0.300 + 3%$0.309

Output · 1M tokens

$2.50 + 3%$2.58

Cache read · 1M tokens

$0.030 + 3%$0.031

Cache write · 1M tokens

$0.083 + 3%$0.086

FEE — FLAT, EVERY MODEL3%

openkey.ai

qwen/qwen3-235b-a22b-2507

Input · 1M tokens

$0.090 + 3%$0.093

Output · 1M tokens

$0.100 + 3%$0.103

FEE — FLAT, EVERY MODEL3%

One workload, priced on both

10M input + 2M output tokens at each model's price, flat 3% fee included.

google/gemini-2.5-flash

$8.24

$8.00 provider + 3%

qwen/qwen3-235b-a22b-2507Cheaper

$1.13

$1.10 provider + 3%

Benchmarks

Design Arena categories where both models have results. Higher Elo and lower rank win.

Gemini 2.5 FlashQwen3 235B A22B Instruct 2507
CategoryEloRankEloRank
3D1148#641071#80
Code1153#691088#81
Data viz1171#581102#77
Game dev1131#721018#94
UI components1148#631022#86
Websites1158#681101#83

Head-to-head preference voting. How we filter and rank

Pricing math

On a 10M input / 2M output token workload: Gemini 2.5 Flash costs $8.00, Qwen3 235B A22B Instruct 2507 costs $1.10 — Gemini's input price is 3.33x higher per token. Provider list price for Gemini is $0.30/M input and $2.50/M output; on OpenKey that's $0.309/M and $2.575/M after the 3% fee ($0.30 x 1.03, $2.50 x 1.03). Qwen3 lists at $0.09/M input and $0.10/M output, becoming $0.0927/M and $0.103/M on OpenKey. Gemini also offers cache pricing ($0.03/M read, $0.083333/M write) that Qwen3 doesn't expose. If you're running high-volume text pipelines, the 7x cost gap on this workload adds up fast; if you need caching to cut repeat-context costs, only Gemini has that lever.

Coding and UI generation performance

Design Arena data is one-sided here. Gemini 2.5 Flash beats Qwen3 235B in every measured category: codecategories (elo 1153, rank 69 vs elo 1088, rank 81), UI components (elo 1148, rank 63 vs elo 1022, rank 86), game dev (elo 1131, rank 72 vs elo 1018, rank 94), website generation (elo 1158, rank 68 vs elo 1101, rank 83), data viz (elo 1171, rank 58 vs elo 1102, rank 77), and 3D (elo 1148, rank 64 vs elo 1071, rank 80). Gemini also has an SVG score (elo 1078, rank 58) that Qwen3 doesn't report. If coding or front-end generation is the core task, Gemini's win rates (43–49% vs Qwen3's 35–48%) are the deciding factor, not price.

Context and modality

Gemini 2.5 Flash handles 1,048,576 tokens of context against Qwen3's 262,144 — a 4x gap. Gemini also accepts text, image, file, audio, and video input; Qwen3 is text-in, text-out only. Max output differs too: Gemini caps at 65,535 completion tokens versus Qwen3's 16,384. For long-document analysis, multi-file review, or any pipeline that needs to ingest audio/video, Gemini is the only option of the two. For pure text-to-text tasks under 262K tokens, the gap doesn't matter.

When to pick each

Pick Gemini 2.5 Flash when the task involves code generation, UI/design work, multimodal input, or context beyond 262K tokens — its `reasoning` parameter support also lets you tune thinking depth per request. Pick Qwen3 235B A22B Instruct 2507 when you're running large-scale text-only workloads (batch summarization, classification, extraction) where the 7x cost advantage compounds across millions of calls. Qwen3's parameter set (top_k, min_p, logit_bias, logprobs) also gives finer sampling control if you're tuning output distributions directly.

Which model for which job

Use casePickWhy
Front-end / UI code generationGemini 2.5 FlashHigher elo and win rate across codecategories, uicomponent, and website Design Arena categories
High-volume text batch jobsQwen3 235B A22B Instruct 2507$1.10 vs $8.00 on a 10M-in/2M-out workload
Long-document or multi-file analysisGemini 2.5 Flash1,048,576 token context vs 262,144, plus file input support
Audio/video input processingGemini 2.5 FlashOnly model of the two with audio and video input modalities
Fine-grained sampling controlQwen3 235B A22B Instruct 2507Supports top_k, min_p, logit_bias, and logprobs parameters Gemini doesn't expose
Repeat-context caching to cut costsGemini 2.5 FlashHas cache_read ($0.03/M) and cache_write ($0.083333/M) pricing; Qwen3 has none

Questions

Which model is cheaper for a typical workload?
Qwen3 235B A22B Instruct 2507 costs $1.10 for 10M input + 2M output tokens versus $8.00 for Gemini 2.5 Flash — roughly 7x cheaper. The gap comes mostly from input pricing: $0.09/M for Qwen3 vs $0.30/M for Gemini, a 3.33x ratio.
Which model has a bigger context window?
Gemini 2.5 Flash supports 1,048,576 tokens versus Qwen3 235B's 262,144 — a 4x difference. That matters for long documents or large codebases where you'd otherwise need to chunk input across multiple calls.
Does either model handle images or audio?
Only Gemini 2.5 Flash does. Its input modalities include text, image, file, audio, and video, while Qwen3 235B A22B Instruct 2507 is text-only in and out. If your pipeline needs to process anything beyond text, Gemini is the only choice here.
Which one is better at coding tasks?
Gemini 2.5 Flash, based on Design Arena's codecategories benchmark: elo 1153 (rank 69, 46.9% win rate) versus Qwen3's elo 1088 (rank 81, 42.7% win rate). Gemini also leads in every other measured category, including UI components and game dev.

Go deeper