Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start free brief

Vals SAGE (SAGE)

Student Assessment with Generative Evaluation

Data verified 24 confirmed releases in the last 30 daysStart free brief

How BenchLM shows SAGE

BenchLM mirrors the public Vals AI SAGE leaderboard captured from https://www.vals.ai/benchmarks/sage and updated by Vals on August 19, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

SAGE is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

Snapshot

77 Vals rows5 task viewsprivate datasetTasks: Overall, Partial Differential Equations, Linear Algebra, Automata, CalculusDisplay only

SAGE score on SAGE — August 19, 2026

We mirror the published sage score view for SAGE. Claude Opus 4.7 leads the public snapshot at 56.10%, followed by Gemma 4 31b It (55.03%) and Claude Opus 4.8 (54.79%). We do not use these results to rank models overall.

77 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated August 19, 2026

SAGE score table (77 models)

Score
1
Claude Opus 4.7Anthropic · Closed
56.10%
2
Gemma 4 31b ItGooglehigh reasoning
55.03%
3
Claude Opus 4.8Anthropic · Closed
54.79%
4
Kimi K3Moonshot AI · Closed
54.26%
5
GPT-5.6 SolOpenAI · Closedmax reasoning
52.56%
6
Qwen3.8-27BAlibaba · Open weightxhigh reasoning
52.40%
7
Claude Opus 4.5 ThinkingAnthropic · Closed
52.09%
8
Claude Fable 5Anthropic · Closed
51.89%
9
Gemini 3 Flash PreviewGooglehigh reasoning
51.85%
10
Claude Opus 4.6 (Adaptive)Anthropic · Closed
51.58%
11
GPT-5.5OpenAI · Closedxhigh reasoning
51.53%
12
Qwen3.8 MaxAlibaba · Open weight
51.25%
13
GPT-5.4 miniOpenAI · Closedxhigh reasoning
50.81%
14
MiniMax M3MiniMax · Open weight
50.57%
15
Kimi K2.6Moonshot AI · Open weight
50.22%
16
Gemini 3.5 FlashGoogle · Closedhigh reasoning
49.88%
17
49.87%
18
Gemini 3.1 Flash Lite PreviewGooglehigh reasoning
49.54%
19
Claude Opus 5Anthropic · Closed
49.43%
20
GPT-5.2OpenAI · Closedxhigh reasoning
49.27%
21
Gemini 3.7 FlashGoogle · Closedhigh reasoning
49.23%
22
Claude Sonnet 5Anthropic · Closed
48.92%
23
Gemini 3.6 FlashGoogle · Closedhigh reasoning
48.69%
24
Gemini 3.1 Pro PreviewGooglehigh reasoning
48.68%
25
Muse Spark 1.2Meta · Closedxhigh reasoning
47.66%
26
Gemini 3 Pro PreviewGooglehigh reasoning
47.62%
27
Gemini 3.5 Flash-LiteGoogle · Closedhigh reasoning
47.30%
28
GPT-5.6 TerraOpenAI · Closedmax reasoning
47.00%
29
Claude Sonnet 4.6Anthropic · Closed
46.58%
30
Muse Spark 1.1Meta · Closedxhigh reasoning
45.58%
31
Qwen3.6-27BAlibaba · Open weight
45.58%
32
Claude Opus 4.5Anthropic · Closed
45.00%
33
Qwen3.6 PlusAlibaba · Closed
44.86%
34
Gemini 2.5 FlashGoogle · Closed
44.76%
35
GPT-5.6 LunaOpenAI · Closedmax reasoning
44.22%
37
GPT-5OpenAIhigh reasoning
43.68%
38
GPT-5.4OpenAI · Closedxhigh reasoning
43.31%
39
MiMo-V2.5Xiaomi · Closed
43.27%
40
GPT-5.1OpenAI · Closedhigh reasoning
43.23%
41
GPT-5 miniOpenAI · Closedhigh reasoning
42.99%
42
Qwen3.5 FlashAlibaba · Closed
42.49%
43
42.10%
44
Gemini 2.5 ProGoogle · Closed
41.92%
45
o3OpenAI · Closedhigh reasoning
41.77%
46
O4 MiniOpenAIhigh reasoning
41.06%
49
Qwen3.7 PlusAlibaba · Closed
39.25%
51
GPT-5.4 nanoOpenAI · Closedhigh reasoning
38.08%
52
Mistral Medium 3.5Mistral AIhigh reasoning
37.61%
53
InklingThinking Machines Lab · Open weight0.99 reasoning
36.55%
54
Claude Sonnet 4.5 ThinkingAnthropic · Closed
36.06%
55
35.70%
56
35.00%
57
Grok 4.5xAI · Closedhigh reasoning
34.97%
59
Claude Sonnet 4.5Anthropic · Closed
32.88%
60
Inkling-SmallThinking Machines Lab · Open weight0.99 reasoning
32.74%
62
Claude Haiku 4.5 ThinkingAnthropic · Closed
31.82%
63
31.38%
64
31.33%
66
30.40%
68
GPT-5 nanoOpenAI · Closedhigh reasoning
30.38%
69
29.76%
70
Grok 4.6xAI · Closedhigh reasoning
28.90%
72
27.53%
73
25.10%
74
24.59%
75
Grok 4.3xAI · Closed
19.74%

The published SAGE snapshot places Claude Opus 4.7 first at 56.10%. The third row is 1.32 points behind. The broader top-10 range is 4.53 points, so many of the published results sit in a relatively narrow band.

77 models have been evaluated on SAGE. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. SAGE is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About SAGE

Year

2026

Tasks

Student assessment with generative evaluation

Format

Accuracy score

Difficulty

Education assessment reasoning

BenchLM mirrors the public Vals AI SAGE leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

BenchLM freshness & provenance

Version

SAGE 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does SAGE measure?

Student Assessment with Generative Evaluation

Which model leads the published SAGE snapshot?

Claude Opus 4.7 currently leads the published SAGE snapshot with 56.10% sage score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on SAGE?

77 AI models are included in BenchLM's mirrored SAGE snapshot, based on the public leaderboard captured on August 19, 2026.

Last updated: August 19, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.