Examenos

92.1% · raw 92.1 %

Headline · max effort · Own vendor

Source/record date: 2026-10-23

Harness: Context compaction at 200K, maximum 500 steps and forced answer at the limit; temperature 1.0/top-p 1.0. Published 23 October 2026; checked 23 October 2026. Actual run dates, serving checkpoint hashes and complete compute budgets are unpublished. Vendor and independently verified results remain separate. K3.1 Pro defaults: max, temperature 1.0, top-p 0.95 unless otherwise stated.

91.7% · raw 91.67 %

Headline · unknown effort · Own vendor

Source/record date: 2026-09-30

Ant’s comparison table on the Ling-3.1-flash card (README commit 95feb31291, 19 October 2026). Effort undisclosed. Re-pointed 2026-11-01 from threatfrontier.com, which transcribed this table; value read from the original.

91.5% · raw 91.5 %

Headline · max effort · Peer vendor

Source/record date: 2026-09-20

Harness: StepFun vendor-reported evaluation; Full agent/grader/task-subset settings unpublished unless stated. Source inspected 8 October 2026; evaluation date unknown unless stated. Approved peer, max effort as labelled; not an independent run.

91.5% · raw 91.5 %

Alternative · max effort · Peer vendor

Source/record date: 2026-10-23

Harness: Context compaction at 200K, maximum 500 steps and forced answer at the limit; temperature 1.0/top-p 1.0. Table labels max with fallback; chart notes mention possible Claude fallbacks/GPT cyberguards. This is not a guaranteed single-model-only execution. Published 23 October 2026; checked 23 October 2026. Actual run dates, serving checkpoint hashes and complete compute budgets are unpublished. Vendor and independently verified results remain separate. As reported by Moonshot; preserve the peer’s own-vendor headline and source-specific setup.

91.2% · raw 91.2 %

Headline · max effort · Own vendor

Source/record date: 2026-07-16

BrowseComp with context-compaction at 300K (90.4 with full 1M / no compaction)

91.2% · raw 91.2 %

Alternative · unknown effort · Peer vendor

Source/record date: 2026-10-05

Harness: Reflection launch evaluation; full agent, revision and token-budget settings not published. As reported by Reflection; peer comparison, not an independent evaluator record. Context management enabled; other agent settings unspecified.

91.2% · raw 91.2 %

Alternative · max effort · Peer vendor

Source/record date: 2026-09-20

Harness: StepFun vendor-reported evaluation; Full agent/grader/task-subset settings unpublished unless stated. Source inspected 8 October 2026; evaluation date unknown unless stated. Approved peer, max effort as labelled; not an independent run. Existing headline retained; alternate vendor configuration/evidence, no averaging.

91.2% · raw 91.2 %

Alternative · unknown effort · Peer vendor

Source/record date: 2026-10-21

Sage-A1.5 report Table 1. As reported by Shanghai AI Laboratory; mixed own-report/best-effort reproduction sources; exact peer effort unreported. Search/visit/Python; top 50 search, summarizer; context reset at 300 calls/context limit, up to 5 rounds. Standard mode without agentic orchestration. Non-headline addition 2026-10-21: preserves the existing own-vendor or earlier configured headline; this source remains reachable as distinct vendor evidence.

91.2% · raw 91.2 %

Alternative · unknown effort · Peer vendor

Source/record date: 2026-10-29

As reported by Mistral in its 29 October chart of BrowseComp; not an independent measurement.

90.8% · raw 90.8 %

Headline · max effort · Peer vendor

Source/record date: 2026-10-23

Harness: Context compaction at 200K, maximum 500 steps and forced answer at the limit; temperature 1.0/top-p 1.0. Table labels max with fallback; chart notes mention possible Claude fallbacks/GPT cyberguards. This is not a guaranteed single-model-only execution. Published 23 October 2026; checked 23 October 2026. Actual run dates, serving checkpoint hashes and complete compute budgets are unpublished. Vendor and independently verified results remain separate. As reported by Moonshot; preserve the peer’s own-vendor headline and source-specific setup.

89.7% · raw 89.7 %

Headline · unknown effort · Own vendor

Source/record date: 2026-09-08

BrowseComp with summary compaction >60% context

88.7% · raw 88.7 %

Headline · high effort · Own vendor

Source/record date: 2026-09-20

Harness: StepFun vendor-reported evaluation; Full agent/grader/task-subset settings unpublished unless stated. Source inspected 8 October 2026; evaluation date unknown unless stated. Step 5 Preview high effort.

87.5% · raw 87.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-07-09

84.5% · raw 84.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-10-29

Mistral’s 29 October chart for the released weights.

83.4% · raw 83.4 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-29

As reported by Mistral in its 29 October chart of BrowseComp; not an independent measurement.

81.7% · raw 81.7 %

Headline · unknown effort · Own vendor

Source/record date: 2026-10-21

Sage-A1.5 report Table 1. Final released checkpoint. Search/visit/Python; top 50 search, summarizer; context reset at 300 calls/context limit, up to 5 rounds. Standard mode without agentic orchestration.

82.8% · raw 82.8 %

Alternative · unknown effort · Own vendor

Source/record date: 2026-10-21

Harness: Agentic Orchestration enabled, spawn_subagent; subagents summarize after 50 turns. Table 4 final model, not teacher. Search/visit/Python; top 50 search, summarizer; context reset at 300 calls/context limit, up to 5 rounds.

76.8% · raw 76.8 %

Headline · unknown effort · Own vendor

Source/record date: 2026-10-17

Harness: Reflection final instruction-tuned BF16; temp=0.7, top_p=0.99. Effort and run date unpublished. Earlier preview results are retained separately.

76.7% · raw 76.7 %

Alternative · unknown effort · Own vendor

Source/record date: 2026-10-17

Harness: Reflection final instruction-tuned FP8; temp=0.7, top_p=0.99. Effort and run date unpublished. Earlier preview results are retained separately.

75.6% · raw 75.6 %

Alternative · unknown effort · Own vendor

Source/record date: 2026-10-17

Harness: Reflection final instruction-tuned NVFP4; temp=0.7, top_p=0.99. Effort and run date unpublished. Earlier preview results are retained separately.

77.4% · raw 77.4 %

Alternative · unknown effort · Own vendor

Source/record date: 2026-10-05

Harness: Reflection launch evaluation; full agent, revision and token-budget settings not published. Beam early-access preview, reported by Reflection. Context management enabled; other agent settings unspecified. Demoted 2026-10-17: retained preview observation; released BF16 card now supplies the current headline.

76.2% · raw 76.2 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-21

Sage-A1.5 report Table 1. As reported by Shanghai AI Laboratory; mixed own-report/best-effort reproduction sources; exact peer effort unreported. Search/visit/Python; top 50 search, summarizer; context reset at 300 calls/context limit, up to 5 rounds. Standard mode without agentic orchestration.

82.5% · raw 82.5 %

Alternative · unknown effort · Peer vendor

Source/record date: 2026-10-29

As reported by Mistral in its 29 October chart of BrowseComp; not an independent measurement.

53.1% · raw 53.1 %

Headline · high effort · Own vendor

Source/record date: 2026-10-20

Harness: SparkLLM own run at thinking high per its card, temperature 1.0, top_p 0.95; exact label “BrowseComp” in the card figure. Self-reported; run date unpublished.

49.2% · raw 49.2 %

Headline · unknown effort · Own vendor

Source/record date: 2026-08-11

Harness: Upstage internal evaluation, common vendor environment; exact harness/version, task subset, sampling and effort unspecified. Exact approved Solar Pro 4; Solar Open 2 is a separate, unapproved comparison peer. Checked 16 October 2026.

46.4% · raw 46.4 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-03

Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). As reported by Aleph Alpha for Qwen3.8 27B; preserves existing own-vendor headline when present.

41.7% · raw 41.7 %

Headline · high effort · Own vendor

Source/record date: 2026-10-19

Harness: SparkLLM own run at thinking high per its card, temperature 1.0, top_p 0.95; exact label “BrowseComp” in the card figure. Self-reported; run date unpublished.

29.4% · raw 29.4 %

Headline · high effort · Own vendor

Source/record date: 2026-10-03

Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101).

10% · raw 10 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-20

SparkLLM-measured peer in its Spark-X2.6 card figure, label “BrowseComp”, not a Google result. Starred peer values on that figure are copied from other cards and are not stored.

8.3% · raw 8.3 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-20

SparkLLM-measured peer in its Spark-X2.6 card figure, label “BrowseComp”, not a Google result. Starred peer values on that figure are copied from other cards and are not stored.

3.7% · raw 3.7 %

Headline · unknown effort · Peer vendor

Source/record date: 2026-10-19

SparkLLM-measured peer in its Spark-X2.6 card figure, label “BrowseComp”, not a Google result. Starred peer values on that figure are copied from other cards and are not stored.