vs
4:2
overall capability wins (0 ties across 6 scored). Agnes-3.0-Flash Preview leads overall.
- Reasoning3:1
- Coding0:1
- Specs1:0
Verdict
Step-3.5 Flash leads coding (widest gap: +2.3 on SciCode); Agnes-3.0-Flash Preview edges reasoning & knowledge; Agnes-3.0-Flash Preview offers larger context (262K vs 256K).
- Step-3.5 Flash leads coding (widest gap: +2.3 on SciCode)
- Agnes-3.0-Flash Preview edges reasoning & knowledge
- Agnes-3.0-Flash Preview offers larger context (262K vs 256K)
- Agnes-3.0-Flash Preview ships open weights (Apache 2.0)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Agnes-3.0-Flash Preview
- -
- Step-3.5 Flash
- $0.16
Price per 1M tokens
Full comparison
Organization
Agnes AI
StepFun
Family
Other
Other
License
Apache 2.0
Proprietary
Open weights
Yes
No
Release
Sep 13, 2026
Feb 18, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
StepFun
Modalities
text, image, video → text
text, image → text
Specs
Context window
262K
256K
Max output
-
16K
Parameters
33B
—
Pricing
Input $/1M
—
$0.09
Output $/1M
—
$0.30
Blended $/1M (3∶1)
-
$0.14
Speed
tok/s
-
120
TTFT (s)
-
0.25
Reasoning
MMLU-Pro
—
—
GPQA Diamond
85.0
83.1
Humanity's Last Exam
—
21.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
23.0
23.6
AA-LCR v1.1
68.3
50.0
CritPt
—
2.5
MMMU-Pro
—
—
IFBench
74.2
64.6
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
—
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
38.1
40.4
CursorBench
—
—
SWE-Rebench
—
59.6
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,394
| Metric | Agnes-3.0-Flash Preview | Step-3.5 Flash | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Agnes AI | StepFun | — |
| Family | Other | Other | — |
| License | Apache 2.0 | Proprietary | — |
| Open weights | Yes | No | — |
| Release | Sep 13, 2026 | Feb 18, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | StepFun | — |
| Modalities | text, image, video → text | text, image → text | — |
| Specs | |||
| Context window | 262K | 256K | A +6K |
| Max output | - | 16K | — |
| Parameters | 33B | — | — |
| Pricing | |||
| Input $/1M | — | $0.09 | — |
| Output $/1M | — | $0.30 | — |
| Blended $/1M (3∶1) | - | $0.14 | — |
| Speed | |||
| tok/s | - | 120 | — |
| TTFT (s) | - | 0.25 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 85.0 | 83.1 | A +2.0 pts |
| Humanity's Last Exam | — | 21.1 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 23.0 | 23.6 | B +0.6 pts |
| AA-LCR v1.1 | 68.3 | 50.0 | A +18.3 pts |
| CritPt | — | 2.5 | — |
| MMMU-Pro | — | — | — |
| IFBench | 74.2 | 64.6 | A +9.6 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 38.1 | 40.4 | B +2.3 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 59.6 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,394 | — |
Agnes-3.0-Flash Preview
- Context
- 262K
- Parameters
- 33B
- Price
- —
- Speed
- —
- Modalities
- text, image, video → text
- License
- Apache 2.0
Agnes AI's (Sapiens AI, Singapore) first open-weight release: a 33B-parameter hybrid-attention model (54 gated delta-rule recurrent layers alternating with 18 global-attention layers, 3:1) under Apache 2.0, with a 262K-token context window and text/image/video input. This open-weight Preview checkpoint is distinct from Agnes AI's production "Agnes 3.0 Flash" API model, which runs a different checkpoint and is not open-weight; the two should not be conflated.
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.