As of 6 Nov 2026, the panel rates Semgrep 3.5/5 and Snyk 2.5/5 in Code Security & Quality (SAST/SCA). Weighted by what each judge cares about, 4 of the 6 judges pick Semgrep and 1 picks Snyk, 1 calls it a tie. Semgrep scores higher on 6 of 7 criteria (Pricing transparency, Developer workflow, quality gates & fixes, Dependencies (SCA), licences & secrets, plus 3 more). Snyk scores higher on 1 of 7 criteria (European sovereignty). Provenance — Semgrep: Rest of world, United States; Snyk: Rest of world. Ratings are the AI panel’s opinion, formed from public evidence: how scores are made.
The short answer
Semgrep leads six of seven criteria: developer workflow 7.8 to 5.7 and dependencies and secrets 7.7 to 5.7 (5 judges to 0, one tie each); code analysis 7.5 to 5.8 (3 Semgrep, 3 tie); integration and administration 6.7 to 5.5 (2 Semgrep, 1 Snyk, 3 tie); trust and compliance 6.7 to 5.7 (2 each way, 2 tie); pricing transparency 5.7 to 3.0 (6 to 0). Snyk leads sovereignty 5.3 to 4.0, 5 judges to 1, its verdict weighing an Irish-law data processing agreement, EU standard contractual clauses and an EU service region against subprocessors including US-hosted OpenAI and Google Cloud Platform; Semgrep's attributes record US legal jurisdiction and non-EU data residency, and its verdict found no public information on EU hosting or an EU contracting entity. Weighted totals: four judges favor Semgrep, one favors Snyk (product security officer, 6.9 to 6.6), one ties at 6.9, and one Snyk total is 0.
Choose Semgrep if
Your team wants findings as pull-request comments with in-thread triage and merge-blocking policies — developer workflow leans Semgrep 5 judges to 0 with one tie (means 7.8 to 5.7), and the verdict records PR and MR comments with in-thread triage and merge-blocking policies.
You need dependency reachability analysis exported to CycloneDX SBOMs — dependencies and secrets leans Semgrep 5 judges to 0 with one tie, and the verdict records reachability analysis for eleven ecosystems written into CycloneDX 1.4-1.7 SBOMs.
You must estimate an annual cost from public pages before a sales cycle — pricing transparency leans Semgrep 6 judges to 0; Snyk's public pages show 1 credit = $1 but no plan or per-developer price, so an annual cost is not computable from them.
Your evaluator matches the engineering manager, platform engineer or skeptic persona — weighted totals favor Semgrep 6.6 to 6.4, 6.8 to 6.3 and 6.3 to 6 respectively.
You rely on automated remediation — the verdict records an autofix that opens a draft pull request in 2 to 10 minutes.
Choose Snyk if
Your procurement requires an EU-law data processing agreement and an EU service region — sovereignty leans Snyk 5 judges to 1 (means 5.3 to 4.0); the verdict weighs an Irish-law data processing agreement, EU standard contractual clauses and an EU service region against a subprocessor chain that includes US-hosted OpenAI and Google Cloud Platform, the latter possibly having access to all customer data.
Your buyer persona is the product security officer — that judge's weighted total favors Snyk 6.9 to 6.6.
You need containers and infrastructure-as-code scanned in the same platform as code and dependencies — Snyk's verdict records coverage of code analysis, open-source dependencies, containers, infrastructure-as-code and secrets.
Your CI runs Jenkins and builds must fail at a configurable minimum severity — the verdict records a Jenkins plugin that fails builds at a configurable minimum severity.
Read this comparison as one judge. Each weighs the same scores by what they care about.
The Skeptic
Has seen "zero false positives" bury a team under four thousand findings on day one, a "free for open source" plan that counted every bot as a contributor, and an "AI autofix" that shipped the repository to a model in the US. Asks how contributors are counted, which languages get real data-flow analysis rather than regular expressions, what happens to the code after a scan, whether the findings export when the contract ends — and what the second year costs.
Semgrep
This judge's pick
Snyk
Criterion by criterion
Code analysis (SAST & code quality)
Semgrep
The depth is real where it counts: cross-file taint analysis is documented per language with a rule language customers write in (pattern-sources, pattern-sinks, propagators, sanitizers), taint traces showing the flow from source to sink, framework-specific control flow for Java, Python, JavaScript and TypeScript, per-language rule counts (Python 710+, JavaScript 250+), and Terraform for infrastructure-as-code. It stays below the top levels because we found no public information on complexity, duplication, coverage import or any code-health trend over time, and the accuracy figures — "triage 80% fewer false positives" and "3.5x more true positives" — are the vendor's own numbers with no independent benchmark on the pages we read.
Snyk
Snyk Code builds an event graph of the source and runs rules over it — framework support is graded Comprehensive or Partial with stated sources-and-sinks coverage and data-flow testing depth, and the Eclipse plugin shows each finding with its data flow and remediation, so this is real analysis rather than pattern matching. Refreshingly, limits are stated plainly: files over 1MB and filenames over 255 characters are skipped. But we found no public information on a customer-writable rule language for code checks, duplication or complexity metrics, or coverage import, and the only published precision figure we found is an 80% fix-accuracy claim with no false-positive rate or independent benchmark for the SAST engine.
Dependencies (SCA), licences & secrets
Semgrep
Reachability analysis is generally available for eleven ecosystems and the documentation states outright that Dart and Elixir have none — a stated limitation I respect — alongside malicious dependency detection, EPSS scoring, CycloneDX SBOMs in versions 1.4 through 1.7 with per-vulnerability reachability state and justification, and licence policies that block or allow with exemptions and named licence sources. Secrets detection covers 630+ credential types with validity checks that run locally on the customer's own infrastructure, pre-commit hooks, and historical scanning explicitly marked Beta; we found no public information on end-of-life component flagging or container image scanning.
Snyk
The dependency tree is built and checked against a proprietary vulnerability database, transitive packages included, with licence compliance, SBOM support, fix pull requests and container image scanning in scope — though the database's sources are not documented, and we found no public information on malicious-package detection or end-of-life components. Secrets blends entropy, regex and ML scoring, reads surrounding code to separate a live credential from a test fixture, soft-blocks commits via pre-commit hooks and gates merges — but we found no public information on scanning the full git history or validity checks against providers, and the scanner states it will not detect fake or AI-generated secrets and caps repositories at 800 findings per scan. Reachability appears as one prioritisation factor rather than a documented per-dependency analysis.
Developer workflow, quality gates & fixes
Semgrep
Findings reach developers where they work: PR and MR comments across GitHub, GitLab, Bitbucket, Jenkins, CircleCI, Azure and Buildkite, triage from the comment thread with a fixed set of documented reasons, remediation policies that can block a merge (the licence check exits with code 1) and control which rules may comment, pre-commit hooks, and an AI autofix that opens a draft pull request in 2 to 10 minutes with its model provider named — Amazon Bedrock, including Anthropic Claude models. What keeps it off the top level: no published autofix acceptance rates, and suppression carries reasons but we found no public information on expiry or audit of suppressions.
Snyk
Pull request checks gate merges, the Jenkins plugin can fail a build with a minimum-severity setting, pre-commit hooks soft-block secrets before they land, and findings can be ignored through an approval workflow with consistent ignores — the workflow basics are real. Fixes arrive as one-click, customisable pull requests backed by the DeepCode AI engine, with Agent Fix automating them. We found no public information on typical scan times, whether IDE results match the server's rules, or deduplication across tools, so the noise story rests on claims like finding unsafe code "up to 50x faster" rather than documented numbers.
Platforms, deployment & administration
Semgrep
On-prem and self-managed repositories are supported, scans can run on customer-controlled infrastructure with a Semgrep Network Broker mediating SCM access, Bitbucket Data Center pull request comments are documented down to the token permissions, and the REST API exists in two versions with per-endpoint maturity labels (Stable, Beta, Experimental) — the kind of candour I like in an API. We found no public information on SCIM provisioning, export of findings in SARIF or CSV, or granular roles per team, and portfolio reporting appears only as "Dashboard and reporting capabilities" with no trends shown.
Snyk
GitHub (including server variants), GitLab, Bitbucket Cloud and Data Center, and Azure Repos are supported, a broker sits between Snyk and the SCM for security and auditability, and roles can be assigned at account, team and target level with multiple roles per user. A REST API is documented alongside V1 and OAuth2 variants, Jira and ServiceNow integrations exist, and detection and remediation trends are reported across every repository for leadership. Three caveats: the majority of Snyk APIs are restricted to Enterprise plan customers only, and we found no public information on single sign-on, SCIM provisioning, audit logs, or SARIF import and export.
Source-code handling, security & compliance
Semgrep
Strong bones: SOC 2 Type II certified with the report period stated (December 1, 2023 to November 30, 2024), a named vendor (Include Security) behind a full-scope whitebox penetration test, two disclosed incidents with the statement that customer findings and source code were not accessed, CLI scans on customer-controlled infrastructure, and AI providers named (OpenAI as default for Multimodal, Amazon Bedrock for autofix) with a per-model opt-out. The caveats a buyer must weigh: we found no public information on code retention periods or a published SLA, Semgrep Inc. is stated as not itself ISO 27001 certified, and the terms grant the company a perpetual, sublicensable licence to create derivative works of Customer Data "to improve and develop Semgrep products and services" — read that clause against the opt-outs before assuming it covers nothing.
Snyk
The terms state plainly that inputs are not used to train, enhance or improve AI models incorporated in the services, ISO 27001, ISO 27017 and a Schellman-audited SOC 2 Type II are certified annually with the report period stated and copies available through the account team, a public status page runs on Atlassian Statuspage, and security incidents are notified within 72 hours. Compliance reporting is mapped to SOC 2, ISO 27001, PCI DSS 4.0, the Cyber Resilience Act and OWASP Top 10. We found no public information on encryption specifics or code retention periods, and deletion on termination falls to the customer with support assistance rather than a stated vendor process.
European sovereignty
Semgrep
The documented chain is American throughout: Semgrep, Inc., a Delaware corporation under California law with exclusive venue in San Francisco County, servers stated to be located in the United States on AWS, and named processors WorkOS for authentication, OpenAI as the default Multimodal model provider, Amazon Bedrock, and Google Analytics. A data processing agreement, a GDPR listing and a subprocessor list do exist in the trust centre, and CLI scans on customer-controlled infrastructure can keep source code off Semgrep's servers. We found no public information on EU hosting, an EU contracting entity or EU residency for the platform where findings and triage are stored.
Snyk
Customers outside the US contract with Snyk Limited, registered in England and Wales, under a DPA governed by Irish law with a Brussels GDPR representative, published EU SCCs, an Irish supervisory authority, and a subprocessor list published with locations, thirty days' notice and objection rights. But the chain matters: the hosting backbone is AWS at US, EU or Australia at the customer's election, while Google Cloud Platform is US-hosted and "may have access to all Customer Data", and the generative AI providers include US-hosted OpenAI and Google Vertex. We found no public information on where the customer's source code and findings are actually stored.
Pricing transparency
Semgrep
The published parts are concrete: the free tier allows up to 10 repositories and a maximum of 10 contributors with Code and Supply Chain at $0/month/contributor and 60 AI credits, and Teams prices each scanner — Code ($30/month/contributor), Supply Chain ($30/month/contributor), Secrets ($15/month/contributor) — with private repository limits of 10, 500 and unlimited by tier and OIDC + SAML from Teams up. The gaps stop me computing a real annual cost: the Enterprise tier is "volume pricing" with no figure, "contributor" is never defined on the pricing page — an undefined contributor is exactly how every bot ends up on an invoice — and we found no public information on monthly versus annual terms or VAT treatment.
Snyk
The model is credits — 1 credit = $1 — with per-capability rates published for exactly two capabilities ("1.0 credits per Active Machine / per Day" for Evo ADS and 4,000 credits per Assessment for AI Pentesting), but no plan price, no per-developer or per-contributor figure, and no counting rule on the captured pages. The second year is not computable: renewals are priced at Snyk's then-current rates and rate card, unused credits expire with the term, fees are non-cancellable and non-refundable, overage is invoiced in arrears at then-current rates, and everything is payable in US dollars exclusive of taxes. A free tier and a no-cost plan for open-source maintainers exist, and small teams get a flat rate whose amount is not stated.
Sovereignty, side by side
Dimension
Semgrep
Snyk
Legal entity
Incorporated in US
Not determined
Hosting location
Outside the EU
Not determined
Hosting options
Self-hostable
Not determined
Open source
Open source (copyleft)
Not determined
Data processing agreement
On request
Published, self-serve
Transfer impact assessment
Not determined
Not determined
Facts, side by side
Only facts both products carry under the same definition — anything else would not be a fair row.
As of 6 Nov 2026, the panel rates Semgrep higher: 3.5/5 against 2.5/5. Weighted by what each judge cares about, 4 of the 6 judges pick Semgrep and 1 picks Snyk, 1 calls it a tie. Which one fits depends on the criteria that matter to you; the table on this page shows both side by side, and each judge’s reading has its own page. Both were scored by the same AI bench under the same rubric, from public evidence only, as the methodology describes.
Is Semgrep or Snyk European?
Semgrep is not European by provenance (Rest of world, United States). Snyk is not European by provenance (Rest of world). Provenance is read from each company’s own legal pages: who builds it and who owns it.
Where do Semgrep and Snyk store customer data?
Per the vendors’ own pages as captured — Semgrep: Outside the EU by default; pages mention the United States. Snyk: its pages as captured do not say.
Do Semgrep and Snyk have a free plan?
As of 6 Nov 2026, both vendors’ own pages state a free plan. Each product page links the pricing page it was read from.