Capability
Unranked
field median 56.2
Not eligible for a public rank
SWE-2
Released Sep 10, 2026 — see all recent releases
Decision reading
SWE-2 is tracked, but not publicly ranked yet. The profile exposes 5 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Data as of September 10, 2026 · How the score is built
Share or export
Strongest published evidence
Published rows are visible, but no category has enough eligible evidence for a comparative rank.
Validate before choosing
5 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Price
API rate not published
input median $1
No comparable first-party hosted token rate
Speed
Not measured
field median 86.5 tok/s
Time to first token not measured
Context
1Mtokens
field median 256,000
Maximum output length is tracked separately
How much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic2/2 verified
- Coding3/3 verified
- ReasoningNot measured
- KnowledgeNot measured
- MathNot measured
- MultilingualNot measured
- MultimodalNot measured
- Inst. FollowingNot measured
Verified sourceProvisionalNot measured
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- Not publishedCognition SWE-2 launch post
- Context window
- 1MCognition SWE-2 launch post
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- Not sourced yet
- Output modalities
- Not sourced yet
- Parameters
- Not disclosed by the provider
- Availability
- Released September 10, 2026 as Cognition's most capable coding model, post-trained from Kimi K3 (2.8T parameters) with reinforcement learning that trains the medium, high, and max effort levels in one run. Available in Devin Desktop and Devin CLI at launch and rolling out to Devin Web and Fusion; Cognition has not published a standalone API, pricing, or weights.
- Cloud regions
- Not tracked yet
- Lifecycle
- Current
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordCognition SWE-2 launch post
- Self-host
- Weights are not published
- Rate limits
- Not tracked yet
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticWeight 22%2 benchmarksVerified | Score pending | Not ranked | Not available | 22% | 2 benchmarks | Verified |
| CodingWeight 20%3 benchmarksVerified | Score pending | Not ranked | Not available | 20% | 3 benchmarks | Verified |
| ReasoningWeight 17%0 benchmarksNot measured | Not measured | Not ranked | Not available | 17% | 0 benchmarks | Not measured |
| KnowledgeWeight 12%0 benchmarksNot measured | Not measured | Not ranked | Not available | 12% | 0 benchmarks | Not measured |
| MathWeight 5%0 benchmarksNot measured | Not measured | Not ranked | Not available | 5% | 0 benchmarks | Not measured |
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | Not ranked | Not available | 7% | 0 benchmarks | Not measured |
| MultimodalWeight 12%0 benchmarksNot measured | Not measured | Not ranked | Not available | 12% | 0 benchmarks | Not measured |
| Inst. FollowingWeight 5%0 benchmarksNot measured | Not measured | Not ranked | Not available | 5% | 0 benchmarks | Not measured |
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding3 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| FrontierCode 1.1 Main | Score50.0% | Versus best verified row Best verified: Claude Fable 5 · 53.5% | Gap3.5 behind | WeightDisplay only | Provider exact |
| Terminal-Bench 2.1Terminal-Bench 2.1 (provider run) | Score92.8% | Versus best verified row Best verified: SWE-2 · 92.8% | GapBest verified | WeightDisplay only | Provider exact |
| DeepSWE | Score73.0% | Versus best verified row Best verified: Muse Spark 1.3 · 75.4% | Gap2.4 behind | WeightDisplay only | Provider exact |
Agentic2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| Terminal-Bench 2.1Terminal-Bench 2.1 (provider run) | Score92.8% | Versus best verified row Best verified: SWE-2 · 92.8% | GapBest verified | WeightDisplay only | Provider exact |
| Terminal-Bench 4.0 | Score27.30% | Versus best verified row Best verified: Claude Mythos 5.1 · 60.90% | Gap33.6 behind | WeightDisplay only | Provider exact |
Lineage
The sequence follows explicit supersedes links. A successor's displayed score stays at least 0.1 points above its predecessor; raw benchmark rows do not move. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Jul 8, 2026
SWE-1.7Not publicly ranked · Price not listed
Sep 10, 2026 · you are here
SWE-2Not publicly ranked · Price not listed
Base entry
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track SWE-2, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
SWE-2 is a proprietary model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Released September 10, 2026 as Cognition's most capable coding model, post-trained from Kimi K3 (2.8T parameters) with reinforcement learning that trains the medium, high, and max effort levels in one run. Available in Devin Desktop and Devin CLI at launch and rolling out to Devin Web and Fusion; Cognition has not published a standalone API, pricing, or weights.
Official exact-value snapshot from Cognition's September 10, 2026 SWE-2 launch post. BenchLM maps only the SWE-2 column of Cognition's coding benchmark table: FrontierCode 1.1 Main, Terminal-Bench 2.1, DeepSWE 1.1, and Terminal-Bench 4. Cognition reports the public result where one exists and otherwise an internal Devin CLI run at the best reasoning-effort setting, so every row is provider-run and harness-specific and the model stays out of weighted ranking. The context window is inherited from the Kimi K3 base model because Cognition does not publish a separate figure. Cognition also reports SWE-2 medium taking 58% fewer turns and costing 81% less than SWE-1.7 on FrontierCode 1.1 Main; those efficiency claims stay descriptive.
SWE-2 sits in the SWE family with SWE-1.7. Its explicit predecessor is SWE-1.7. 5 of 434 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Radar
Full release historySWE-2 release history
SWE-2 released
Cognition · Model release
Source confirmed
Radar confirmed these at the source. Use SWE-2 in your work? Explore Radar to follow supported changes and choose your alerts.
Frequently asked questions
How does SWE-2 perform overall in AI benchmarks?
SWE-2 has 5 source-displayable benchmark rows, but it does not qualify for a public overall rank. The available rows remain visible by category without being converted into a site-wide score. Missing evidence stays blank instead of being estimated from an earlier model.
Is SWE-2 good for coding and programming?
SWE-2 has source-displayable benchmark coverage for coding and programming, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is SWE-2 good for agentic tool use and computer tasks?
SWE-2 has source-displayable benchmark coverage for agentic tool use and computer tasks, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Which sibling models are related to SWE-2?
SWE-2 belongs to the SWE family. Related tracked variants include SWE-1.7. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.
Does SWE-2 have full benchmark coverage on BenchLM?
No. SWE-2 currently has 5 source-displayable rows across 434 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of SWE-2?
SWE-2 has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.
Compare SWE-2 with every tracked model482 comparisons
Last updated September 10, 2026. Runtime fields remain blank until a sourced snapshot exists.
Watch SWE-2 in the weekly brief
Get one weekly email when material rank, price, availability, or benchmark evidence changes are worth revisiting.
Read a sample issueJoin 2,000+ readers.