vs
1:7
overall capability wins (0 ties across 8 scored). Step-3.5 Flash leads overall.
- Reasoning1:5
- Coding0:1
- Pricing0:3
- Specs0:1
Verdict
Step-3.5 Flash leads coding (widest gap: +2.7 on SciCode); Step-3.5 Flash offers ~2.0x larger context (256K vs 128K); Step-3.5 Flash is ~2.6x cheaper on a blended token basis than Mercury 2.
- Step-3.5 Flash leads coding (widest gap: +2.7 on SciCode)
- Step-3.5 Flash offers ~2.0x larger context (256K vs 128K)
- Step-3.5 Flash is ~2.6x cheaper on a blended token basis than Mercury 2
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Mercury 2
- $0.44
- Step-3.5 Flash
- $0.16
Step-3.5 Flash is about 2.7x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Inception Labs
StepFun
Family
Other
Other
License
Proprietary
Proprietary
Open weights
No
No
Release
Sep 1, 2026
Feb 18, 2026
Knowledge cutoff
-
-
API / provider
Inception Labs
StepFun
Modalities
text → text
text, image → text
Specs
Context window
128K
256K
Max output
-
16K
Parameters
—
—
Pricing
Input $/1M
$0.25
$0.09
Output $/1M
$0.75
$0.30
Blended $/1M (3∶1)
$0.38
$0.14
Speed
tok/s
-
120
TTFT (s)
-
0.25
Reasoning
MMLU-Pro
—
—
GPQA Diamond
77.0
83.1
Humanity's Last Exam
17.1
21.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
21.2
23.6
AA-LCR v1.1
43.7
50.0
CritPt
0.8
2.5
MMMU-Pro
—
—
IFBench
69.8
64.6
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
27.3
—
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
37.7
40.4
CursorBench
—
—
SWE-Rebench
—
59.6
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,167
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,394
| Metric | Mercury 2 | Step-3.5 Flash | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Inception Labs | StepFun | — |
| Family | Other | Other | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Sep 1, 2026 | Feb 18, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Inception Labs | StepFun | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 256K | B +128K |
| Max output | - | 16K | — |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $0.25 | $0.09 | B +$0.16 |
| Output $/1M | $0.75 | $0.30 | B +$0.45 |
| Blended $/1M (3∶1) | $0.38 | $0.14 | B +$0.23 |
| Speed | |||
| tok/s | - | 120 | — |
| TTFT (s) | - | 0.25 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 77.0 | 83.1 | B +6.1 pts |
| Humanity's Last Exam | 17.1 | 21.1 | B +4.0 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 21.2 | 23.6 | B +2.4 pts |
| AA-LCR v1.1 | 43.7 | 50.0 | B +6.3 pts |
| CritPt | 0.8 | 2.5 | B +1.7 pts |
| MMMU-Pro | — | — | — |
| IFBench | 69.8 | 64.6 | A +5.2 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 27.3 | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 37.7 | 40.4 | B +2.7 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 59.6 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,167 | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,394 | — |
Mercury 2
- Context
- 128K
- Parameters
- —
- Price
- $0.25 / $0.75
- Speed
- —
- Modalities
- text → text
- License
- Proprietary
Inception Labs' diffusion-based (non-autoregressive) reasoning LLM, API-only with a 128K context window and high-throughput inference on Blackwell GPUs.
Step-3.5 Flash
- Context
- 256K / 16K out
- Parameters
- —
- Price
- $0.09 / $0.30
- Speed
- 120 tok/s · 0.25s TTFT
- Modalities
- text, image → text
- License
- Proprietary
StepFun's fast multimodal API tier — competitive STEM scores at Flash pricing.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.