Gemma 4 31B IT vs MiMo-V2.5 vs Step 3.5 Flash 2603
MiMo-V2.5 comes out ahead, 79 to 70 and 62 on our weighted score, though Step 3.5 Flash 2603 is 14% cheaper per token.
GoogleGemma 4 31B IT
70/100- ECI142.8
- Price$0.14 / $0.40
- Context262K
- Our pick
XiaomiMiMo-V2.5
79/100- ECI—
- Price$0.14 / $0.28
- Context1.05M
StepFunStep 3.5 Flash 2603
62/100- ECI—
- Price$0.10 / $0.30
- Context256K
MiMo-V2.5 is our pick
MiMo-V2.5 is the better all-round choice, scoring 79/100 against Gemma 4 31B IT (70) and Step 3.5 Flash 2603 (62). It leads on inputs & features and context window. Step 3.5 Flash 2603 wins on 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 priceStep 3.5 Flash 2603Step 3.5 Flash 2603 $0.15 · MiMo-V2.5 $0.175 · Gemma 4 31B IT $0.205 per 1M tokens (3:1 blend)
- Longest contextMiMo-V2.5MiMo-V2.5 1,048,576 · Gemma 4 31B IT 262,144 · Step 3.5 Flash 2603 256,000 tokens
- Widest inputsMiMo-V2.5Gemma 4 31B IT: Text, Images · MiMo-V2.5: Text, Images, Audio, Video · Step 3.5 Flash 2603: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | Gemma 4 31B IT | MiMo-V2.5 | Step 3.5 Flash 2603 |
|---|---|---|---|---|
| Price | 50% | 83 | 86 | 89 |
| Inputs & features | 30% | 70 | 80 | 35 |
| Context window | 20% | 37 | 61 | 36 |
| Overall | 100% | 70/100 | 79/100 | 62/100 |
Left out because at least one model lacks the data: capability. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | Gemma 4 31B ITGoogle | MiMo-V2.5Xiaomi | Step 3.5 Flash 2603StepFun |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 142.8 | — | — |
| ECI rank | #86 of 148 | — | — |
| GPQA DiamondGraduate-level science questions | 75.8% | — | — |
| OTIS Mock AIME 2024–2025Competition mathematics | 73.3% | — | — |
| SimpleQA VerifiedShort factual questions | 10.4% | — | — |
| Price per million tokens | |||
| Input | $0.14 | $0.14 | $0.10 (best) |
| Output | $0.40 | $0.28 (best) | $0.30 |
| Cached input | — | $0.0028 (best) | $0.02 |
| Blended (3:1) | $0.205 | $0.175 | $0.15 (best) |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 30 providers | Official Xiaomi API | Official StepFun (Global) API |
| Limits | |||
| Context window | 262,144 tokens | 1,048,576 tokens (best) | 256,000 tokens |
| Max output | 32,768 tokens | 131,072 tokens | 256,000 tokens (best) |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | Yes | Yes | No |
| PDFs | No | No | No |
| Audio | No | Yes | No |
| Video | No | Yes | No |
| Reasoning | Yes | Yes | Yeslow · high |
| Tool calling | Yes | Yes | Yes |
| Structured output | Yes | No | No |
| Availability | |||
| Weights | Open | Open | Open |
| API model ID | gemma-4-31b-it | mimo-v2.5 | step-3.5-flash-2603 |
| API providers | 38 (best) | 21 | 3 |
| Released | Apr 2, 2026 | Apr 22, 2026 | Apr 2, 2026 |
| Knowledge cutoff | — | Dec 2024 | Jan 2025 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- Gemma 4 31B IT$2.20
- MiMo-V2.5$1.96
- Step 3.5 Flash 2603$1.60
Which should you choose?
Which is better: Gemma 4 31B IT, MiMo-V2.5 or Step 3.5 Flash 2603?
MiMo-V2.5 is the better all-round choice, scoring 79/100 against Gemma 4 31B IT (70) and Step 3.5 Flash 2603 (62). It leads on inputs & features and context window. Step 3.5 Flash 2603 wins on 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, Gemma 4 31B IT, MiMo-V2.5 or Step 3.5 Flash 2603?
Step 3.5 Flash 2603 is cheaper at $0.10 input / $0.30 output per million tokens (official StepFun (Global) API price). MiMo-V2.5 costs $0.14 input / $0.28 output per million tokens (official Xiaomi API price); Gemma 4 31B IT costs $0.14 input / $0.40 output per million tokens (median across 30 API providers). At a typical mix of three input tokens to one output token, that is $0.15 per million tokens for Step 3.5 Flash 2603 versus $0.175 for MiMo-V2.5 (1.2× as much) and $0.205 for Gemma 4 31B IT (1.4× as much).
Which scores higher on benchmarks?
There is no independent benchmark that covers all three models yet. Gemma 4 31B IT has an ECI of 142.8, MiMo-V2.5 has not been scored yet and Step 3.5 Flash 2603 has not been scored yet.
Which is better for coding?
There are no published SWE-bench Verified results for Gemma 4 31B IT, MiMo-V2.5 and Step 3.5 Flash 2603 yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.
Which has the bigger context window?
MiMo-V2.5 has the largest context window at 1,048,576 tokens, against 262,144 for Gemma 4 31B IT and 256,000 for Step 3.5 Flash 2603. Maximum output per response: Gemma 4 31B IT up to 32,768, MiMo-V2.5 up to 131,072, Step 3.5 Flash 2603 up to 256,000 tokens.
Which can read images, PDFs, audio or video?
Gemma 4 31B IT accepts text and images; MiMo-V2.5 accepts text, images, audio and video; Step 3.5 Flash 2603 accepts text. MiMo-V2.5 handles the widest range of inputs.
Are any of these open source?
Yes, all three publish their weights, so you can self-host them.
Which is newer?
MiMo-V2.5 is the newest, released Apr 22, 2026. Gemma 4 31B IT came out Apr 2, 2026; Step 3.5 Flash 2603 came out Apr 2, 2026. Knowledge cutoff: MiMo-V2.5 Dec 2024, Step 3.5 Flash 2603 Jan 2025.
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 Oct 4, 2026.
