State of LLM eval tools and benchmarks, October 2026
Last updated 25 Oct 2026
Short answer: Evals Review tracks 122 AI evals and benchmarks across 11 categories and 12 LLM evaluation platforms and frameworks; of the 79 evals whose licence we checked, 73 (92%) use MIT or Apache-2.0. 6 of the 12 tools publish their main code on GitHub, and 4 eval or observability tools have shut down since 2025, with OpenAI's hosted Evals scheduled to close on 30 Nov 2026.
Key numbers
- 122 evals and benchmarks in 11 categories.
- 79 have a checked licence; 73 of them (92%) are MIT or Apache-2.0.
- 95 list a dated last update (repo commit, release or publication).
- 12 eval platforms and frameworks tracked (9 platforms, 3 frameworks); 6 publish their main code on GitHub.
- 15 open-source eval repositories with a GitHub star count taken on 27 Oct 2026.
- 4 tools shut down since 2025, plus 1 scheduled closure.
Evals by category
| Category | Evals | Share |
|---|---|---|
| Agents & tool use | 22 | 18% |
| Coding | 21 | 17% |
| Reasoning & knowledge | 18 | 15% |
| Rubric libraries & frameworks | 14 | 11% |
| Safety & factuality | 13 | 11% |
| Image & vision | 8 | 7% |
| Writing & instruction following | 7 | 6% |
| RAG & retrieval | 7 | 6% |
| Healthcare | 6 | 5% |
| Multilingual & Indian languages | 4 | 3% |
| Customer support | 2 | 2% |
Licence mix of evals
| Licence group | Evals |
|---|---|
| Not checked | 43 |
| MIT | 37 |
| Apache-2.0 | 36 |
| Creative Commons | 3 |
| Other (GPL, Elastic, research-only) | 3 |
Open-source eval tools by GitHub stars
| Project | Repository | Licence | Stars | Latest release |
|---|---|---|---|---|
| Langfuse | langfuse/langfuse | MIT core; ee/ folders under a separate licence | 36,562 | v4.70.1 (26 Oct 2026) |
| MLflow | mlflow/mlflow | Apache-2.0 | 28,549 | v3.17.2 (21 Oct 2026) |
| Promptfoo | promptfoo/promptfoo | MIT | 26,480 | 0.125.1 (25 Oct 2026) |
| Comet Opik | comet-ml/opik | Apache-2.0 | 22,952 | 2.2.121 (26 Oct 2026) |
| DeepEval | confident-ai/deepeval | Apache-2.0 | 19,129 | python-v4.2.4 (22 Sept 2026) |
| Ragas | vibrantlabsai/ragas | Apache-2.0 | 16,160 | v0.4.3 (13 Jan 2026) |
| LM Evaluation Harness | EleutherAI/lm-evaluation-harness | MIT | 14,323 | v0.4.13 (31 Aug 2026) |
| Arize Phoenix | Arize-ai/phoenix | Elastic License 2.0 | 12,017 | arize-phoenix-v20.26.1 (25 Oct 2026) |
| Helicone | Helicone/helicone | Apache-2.0 | 6,240 | v2025.08.21-1 (21 Aug 2025) |
| simple-evals | openai/simple-evals | MIT | 4,675 | none published |
| Laminar | lmnr-ai/lmnr | Apache-2.0 | 3,451 | v0.3.2 (16 Oct 2026) |
| HELM | stanford-crfm/helm | Apache-2.0 | 2,960 | v0.5.16 (30 Apr 2026) |
| OpenLIT | openlit/openlit | Apache-2.0 | 2,887 | openlit-2.1.1 (26 Oct 2026) |
| Autoevals (Braintrust) | braintrustdata/autoevals | MIT | 1,069 | py-0.4.1 (21 Oct 2026) |
| Inspect Evals | UKGovernmentBEIS/inspect_evals | MIT | 720 | v0.26.0 (23 Oct 2026) |
Tools shut down in 2025-26
| Tool | What its own notice says | Source |
|---|---|---|
| Humanloop | Platform sunset 8 Sep 2025 | Humanloop migration guide |
| Literal AI | Service available until 31 Oct 2025, then discontinued | Literal AI docs |
| Gentrace | Homepage says it has shut down; code on GitHub under MIT (no date on page) | gentrace.ai |
| WhyLabs | Homepage says the company is discontinuing operations and the platform is open-sourced (no date on page) | whylabs.ai |
| OpenAI hosted Evals | Scheduled: read-only 31 Oct 2026, dashboard and API shut down 30 Nov 2026 | OpenAI deprecations |
How we compiled this
- Every number is computed when the site is built, from the same data files behind our eval and tool pages. Nothing is estimated.
- Evals: each entry's creator, category, licence and last update come from its paper, repository or official page (links first checked 15 Oct 2026). "Not checked" means we haven't verified the licence.
- Tools: features and prices were read on each vendor's own pages on 27 Oct 2026. GitHub stars and releases came from the GitHub API on 27 Oct 2026.
- Shut-downs: listed only when the vendor's own site or docs say so.
- Nothing here is ranked, and no tool paid to be included. See how we review and our disclosure.
Cite this page
Evals Review. "State of LLM eval tools and benchmarks (October 2026)." evalsreview.com, 25 Oct 2026. https://evalsreview.com/state-of-llm-evals
FAQ
How many AI evals and benchmarks does Evals Review track?
122 as of 25 Oct 2026, in 11 categories. The largest are Agents & tool use (22) and Coding (21).
Which licence do most AI benchmarks use?
Of the 79 evals whose licence we checked, 37 use MIT and 36 use Apache-2.0. 43 are not checked yet.
Which open-source LLM eval tool has the most GitHub stars?
On 27 Oct 2026 it was Langfuse (langfuse/langfuse) with 36,562 stars, from the GitHub API. Stars change daily.
Which LLM eval tools have shut down?
Humanloop (sunset 8 Sep 2025), Literal AI (discontinued after 31 Oct 2025), Gentrace and WhyLabs (their homepages say so, without a date). OpenAI's hosted Evals is scheduled to shut down on 30 Nov 2026.
Can I reuse this data?
Yes. The dataset is CC BY 4.0. Credit Evals Review and link to this page or evalsreview.com/data.
How often is this updated?
We re-check the data when we add evals or re-read vendor pages. Each figure on this page shows the date it was checked.
Are the tools ranked?
No. Tables are sorted by count or by GitHub stars only. Stars measure attention, not quality.