Compare
Four models, one page
Pick the models you are actually choosing between. Every row is scaled within itself, so a bar means something next to its neighbours and nothing across rows.
| Specification | Step 3.7 FlashStepFun | Claude Opus 5Anthropic |
|---|---|---|
| Released | May 28, 2026 | Jul 24, 2026 |
| Context window | 262K | 1M |
| Max output | 256K | 128K |
| Input / 1M | $0.2 | $5 |
| Output / 1M | $1.15 | $25 |
| Cost per task | $0.10 | $2.25 |
| Arena Elo | 1,211 | 1,393 |
| Serving providers | 3 | 5 |
| Parameters | 201.4B | Undisclosed |
| Licence | apache-2.0 | Proprietary |
| Capabilities |
|
|
Metrics side by side
Each row is scaled to the largest value in that row — bars compare within a row, never across rows
- Step 3.7 Flash
- Claude Opus 5
Intelligence Index
Step 3.7 Flash30.9Claude Opus 563.1Coding Index
Step 3.7 Flash39.6Claude Opus 578.0Agentic Index
Step 3.7 Flash21.7Claude Opus 559.2Output speed
Step 3.7 Flash153 t/sClaude Opus 587 t/sContext window
Step 3.7 Flash262KClaude Opus 51MLatency · lower is better
Step 3.7 Flash253msClaude Opus 51.12sBlended price / 1M · lower is better
Step 3.7 Flash$0.438Claude Opus 5$10Cost per task · lower is better
Step 3.7 Flash$0.10Claude Opus 5$2.25
Rows marked “lower is better” still draw a longer bar for a larger number — read the value, not just the length. Arena Elo is in the specification table above instead: it has no meaningful zero, so a bar would flatten the gaps.
View as table
| Metric | Step 3.7 Flash | Claude Opus 5 |
|---|---|---|
| Intelligence Index | 30.9 | 63.1 |
| Coding Index | 39.6 | 78.0 |
| Agentic Index | 21.7 | 59.2 |
| Output speed | 153 t/s | 87 t/s |
| Context window | 262K | 1M |
| Latency · lower is better | 253ms | 1.12s |
| Blended price / 1M · lower is better | $0.438 | $10 |
| Cost per task · lower is better | $0.10 | $2.25 |
Evaluation scores
Percentage correct on a common 0–100% scale
- Step 3.7 Flash
- Claude Opus 5
GPQA Diamond
Step 3.7 Flash80.9%Claude Opus 593.2%Humanity's Last Exam
Step 3.7 Flash21.4%Claude Opus 554.9%SciCode
Step 3.7 Flash40.0%Claude Opus 555.7%τ²-bench
Step 3.7 Flash98.5%Claude Opus 542.1%Terminal-Bench Hard
Step 3.7 Flash35.6%Claude Opus 5not measuredAA-LCR long context
Step 3.7 Flash69.7%Claude Opus 575.7%
A missing bar means that evaluation was not run for that model — it is not a zero.
View as table
| Evaluation | Step 3.7 Flash | Claude Opus 5 |
|---|---|---|
| GPQA Diamond | 80.9% | 93.2% |
| Humanity's Last Exam | 21.4% | 54.9% |
| SciCode | 40.0% | 55.7% |
| τ²-bench | 98.5% | 42.1% |
| Terminal-Bench Hard | 35.6% | — |
| LiveCodeBench | — | — |
| AA-LCR long context | 69.7% | 75.7% |
| AIME 2025 | — | — |
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).