Hy4 preview vs SWE-2: Features, Pricing & Which Is Better (2026)

A side-by-side comparison of Hy4 preview and SWE-2 — features, pricing, and ideal use cases — to help you decide which AI tool fits your workflow.

Hy4 preview

Tencent

Free

Tencent's open-weight Hy4 preview, a 770B-parameter Mixture-of-Experts model with 49B active parameters and a 1M-token context window.

770B Mixture-of-Experts Architecture: Holds 770 billion total parameters while activating only 49 billion per token, so capacity scales without proportional inference cost.
1M-Token Context Window: Accepts inputs exceeding one million tokens, allowing whole codebases, long document sets or extended agent traces in a single prompt.
Apache 2.0 Open Weights: Released under a permissive licence that allows commercial use, modification and redistribution with no separate agreement.
Productivity Task Focus: Tuned for real-world coding, office work and scientific research rather than narrow benchmark optimisation.
Multi-Product Availability: Accessible globally through Tencent's WorkBuddy, CodeBuddy, Yuanbao and ima applications in addition to the raw weights.
API Access via TokenHub and OpenRouter: Can be called through Tencent Cloud TokenHub or OpenRouter for teams that prefer hosted inference over self-hosting.

Whole-Repository Code Work: Load an entire codebase into the million-token context to reason about refactors and cross-file dependencies at once.
Long-Horizon Agent Tasks: Drive multi-step agent workflows where the full history of tool calls and intermediate results must stay in context.
Self-Hosted Deployment: Run a frontier-scale open-weight model on private infrastructure where data cannot leave the organisation.
Scientific Literature Analysis: Ingest large collections of papers or experimental logs and synthesise findings without chunking the input.
Office Document Processing: Summarise, draft and restructure long reports, contracts and spreadsheets in enterprise workflows.
Commercial Fine-Tuning: Adapt the weights for a proprietary product under the Apache 2.0 licence without negotiating a model licence.

Cognition

Paid

Cognition's coding model that scores 50.0% on FrontierCode 1.1 Main at 64% lower cost than comparable frontier models.

Pareto-Frontier Cost Efficiency: Matches GPT-5.6 Sol and Fable 5/5.1 on coding benchmarks at a fraction of their price and comes within a few points of GPT-6 Astra at roughly a quarter of the cost.
Single-Run Multi-Effort RL: A reinforcement learning algorithm trains all reasoning-effort levels in one run, applying a per-level linear cost penalty derived from the base model's local frontier slope.
Focused Codebase Exploration: Stronger engineering judgment lets the model decide which parts of a repository matter, cutting mean steps per run from 127 to 53 at medium effort.
Selectable Effort Levels: Ships medium, high and max reasoning settings so teams can trade additional steps and cost for accuracy on harder tasks.
End-to-End Test Writing: Produces tests that validate an implementation end to end, catching regressions and edge cases more reliably than previous SWE models.
Resourceful Task Recovery: When an expected route is blocked — an unavailable MCP integration, for example — it finds an alternative path to the same answer within the user's stated boundaries.
Efficient Training and Serving Stack: NVFP4/FP8 kernels, quantization-aware training and an online draft model cut memory use and train-inference mismatch despite nearly 3x the base parameters of SWE-1.7.
Hardened Verifier Flywheel: Triples the number of RL environments, adds instruction-following overlays, and uses earlier SWE-2 checkpoints to iteratively strengthen verifiers.

Agentic Software Engineering: Powering Devin sessions that plan, edit, build and test changes across a real repository with minimal supervision.
Cost-Sensitive Coding at Scale: Teams running large volumes of automated coding tasks pick a model that holds frontier-adjacent accuracy at a materially lower per-task cost.
Terminal and Tooling Workflows: Strong Terminal-Bench results suit tasks driven through shell commands, build systems and command-line tooling.
Regression Test Generation: Generating end-to-end tests for existing implementations to catch edge cases before a release.
Effort-Tiered Task Routing: Routing simple tickets to medium effort and hard migrations to high or max effort within the same model deployment.
Benchmark and Model Evaluation: Engineering leaders compare coding model options on published FrontierCode, DeepSWE and Terminal-Bench numbers alongside cost.