LM Council is a multi-model workspace that lets you orchestrate simultaneous responses from GPT-5, Claude Opus 4, Gemini 2.5, Grok 4, and other frontier AI systems inside a single conversation.

What are Projects and how do they work?

Projects are topic workspaces that group chats and store reusable context. Project notes and up to 25 project files are included in project replies, and saved model slots can act as the project default models. Adding a model reply or chat summary saves it as a project file for future context.

How do I customize which models participate?

You can add, replace, or remove models from each slot, letting you mix GPT-5, Claude Opus 4, Gemini 2.5 Pro, Grok 4, O3, and other providers in one coordinated exchange.

Can I direct a prompt to a specific model?

Use the at-mention picker to target individual models. If no model is selected, every member of the council responds and you can compare their reasoning side-by-side.

How do shared conversations work?

Any transcript can be saved as a read-only link, making it easy to showcase comparisons or hand off results to teammates without exposing private data. When someone creates and verifies an account directly from a page you shared, you also progress toward Rewards like custom emoji defaults and the Mystery Model category.

What rewards can I unlock by sharing?

Sharing chats or code previews that lead to a verified signup unlocks new perks: one verification lets you set the default emoji on the composer button, and five verifications unlock the gold Mystery Model category that loads a random model you can reveal on demand.

What plans are available?

A free tier provides community models and baseline credits, while Plus ($8/month, 350K credits), Pro ($14/month, 700K credits), and MAX ($49/month, 2.8M credits) unlock premium systems including GPT-5, Claude Opus 4, Gemini 2.5, automation features, and higher monthly allowances.

What is the Leader feature?

Leader mode assigns one model to synthesize the council, producing a consolidated summary or recommendation once every other model has finished responding.

What are Credit Blocks?

Credit Blocks are one-time credit purchases available to Plus, Pro, and MAX subscribers. Four packs are offered: Spark (500K credits, $18), Charge (1M credits, $35), Surge (5M credits, $170), and Vault (10M credits, $330). Purchased credits are added to a reserve balance that is consumed only after your monthly allowance is exhausted. Reserve credits never expire while subscribed; if you cancel, a 30-day grace period applies before credits are frozen until re-subscription or a new Credit Block purchase.

Why do Past Chats and Audio sometimes show only recent items first?

Past Chats now loads recent history first for faster app startup. You can press Load older chats (or Load older chats for audio) to continue scanning older items, and chat search can continue finding matches as older pages load.

Which AI model is best for coding in 2026?

In the source snapshots shown here, Claude Opus 4.7 leads SWE-bench Verified at 83.5%, followed by GPT-5.5 at 80.6% and Gemini 3.5 Flash at 79.3%. Claude Opus 4.7 also leads the Terminal-Bench snapshot at 90.2%. Check each card's snapshot and source-checked dates before comparing results.

What is the best AI model for mathematics?

On Epoch AI math benchmarks, GPT-5.5 Pro and GPT-5.5 both reach 100.0% on OTIS Mock AIME 2024-2025. On FrontierMath v2, GPT-5.5 Pro leads Tiers 1-3 at 87.7%, while Claude Fable 5 leads Tier 4 at 87.8%. You can compare these models directly on LM Council.

How do GPT-5 models and Claude compare?

GPT-5.5 Pro leads FrontierMath Tiers 1-3, while GPT-5.4 Pro leads GPQA Diamond. Claude Fable 5 leads FrontierMath Tier 4, Claude Opus 4.7 leads SWE-bench Verified, and Claude Opus 4.8 leads GSO. Results come from separate benchmark snapshots, so compare their dates and methods as well as their scores.

What are the most reliable AI benchmarks?

This page includes standardized benchmarks from Epoch AI and Scale AI, including Humanity's Last Exam (2,500 expert-level questions), FrontierMath v2 (research-level math), GPQA Diamond (PhD-level science), SWE-bench Verified (real-world coding), and METR Time Horizons (long-task endurance).

Can I test these benchmark-winning models myself?

Yes. LM Council provides access to GPT-5.6, Claude Fable 5, Claude Opus 5, Gemini 3, and other top-performing model families. You can run available models together to compare outputs or use this page to compare their benchmark snapshots.

Which AI model is best for long documents?

Fiction.liveBench reports length-specific long-context results, and rankings can change as models and test sets change. Use the benchmark card's snapshot date and full-results link before choosing a model for long documents.

How is GPT-5 different from GPT-4?

GPT-5-class models show large gains over GPT-4-era systems, including 100.0% on OTIS Mock AIME, 87.7% on FrontierMath Tiers 1-3 v2 for GPT-5.5 Pro, and 80.6% on SWE-bench Verified for GPT-5.5. Compare them directly on LM Council to see the difference in real-world applications.

What is Epoch AI and why trust these benchmarks?

Epoch AI is an independent research organization that maintains standardized AI benchmark data. Its public benchmark hub tracks dozens of distinct benchmarks covering mathematics, science, coding, and real-world tasks, with published sources and update dates.

Which models are best for science and reasoning?

On GPQA Diamond, a PhD-level science benchmark, GPT-5.4 Pro leads at 94.6%, followed by Gemini 3.1 Pro Preview at 94.1%, GPT-5.5 at 94.0%, and GPT-5.5 Pro at 93.9%. For common-sense reasoning on SimpleBench, Claude Fable 5 leads at 81.9%.

How can I compare specific AI models across benchmarks?

Use the interactive comparison tool on this page to select any two models, such as GPT-5.6 Sol and Claude Opus 5, and see side-by-side performance across the listed snapshots. You can also create a council with available models to test them on your own prompts.

Are these benchmark results updated regularly?

The page is manually curated. Each benchmark card shows when its result snapshot was captured and when its linked source was last checked. The footer shows only the newest snapshot date across all cards.

What is Humanity's Last Exam?

Humanity's Last Exam (HLE) is a 2,500-question benchmark created by Scale AI and the Center for AI Safety, featuring expert-level questions across mathematics, humanities, and natural sciences. In the linked source snapshot, Gemini 3.1 Pro Preview scores 46.44% and GPT-5.4 Pro scores 44.32%; Scale groups both at rank 1 because its ranking uses confidence intervals.

AI Model Benchmarks

18 source-dated benchmark snapshots, curated by AI Explained, author of SimpleBench

Evaluation settings and metrics vary; use each card's linked source before comparing scores.

Compare Models

Model 1:

Model 2:

Search:

Humanity's Last Exam

Results snapshot Jun 14, 2026 • Source checked Aug 18, 2026

2,500 of the toughest, subject-diverse, multi-modal questions designed to test for both depth of reasoning and breadth of knowledge. Created in partnership with the Center for AI Safety, HLE includes questions across mathematics, humanities, and natural sciences from nearly 1,000 expert contributors.

	Models (no tools)	Score
1	Gemini 3.1 Pro Preview (high thinking)	46.44% ±1.96
1	GPT-5.4 Pro	44.32% ±1.95
3	Muse Spark	40.56% ±1.92
3	Gemini 3 Pro Preview	37.52% ±1.90
4	GPT-5.4 (xhigh)	36.24% ±1.88

	Model	Score
1	Claude Fable 5	81.9%
2	Claude Opus 5	80.6%
3	Gemini 3.1 Pro Preview	79.6%
4	GPT-5.5 Pro	76.9%
5	Gemini 3.5 Flash	76.7%

	Model	Minutes
1	Claude Mythos Preview (early)	1044.8 ±1397.7
2	Claude Opus 4.6 (unknown settings)	718.8 ±1815.2
3	Gemini 3.1 Pro Preview	384.1 ±230.6
4	GPT-5.2 (high)	352.2 ±335.5
5	GPT-5.3 Codex	349.5 ±333.1

	Model	Score
1	Claude Opus 4.7 (max)	83.5% ±1.7
2	GPT-5.5 (xhigh)	80.6% ±1.8
3	Gemini 3.5 Flash (high)	79.3% ±1.8
4	Claude Opus 4.6 (no thinking)	78.7% ±1.9
5	GLM 5.2 (max)	78.7% ±1.9

	Model	Score
1	GPT-5.4 Pro (xhigh)	94.6% ±1.6
2	Gemini 3.1 Pro Preview	94.1% ±1.7
3	GPT-5.5 (xhigh)	94.0% ±1.5
4	GPT-5.5 Pro (xhigh)	93.9% ±1.6
5	GPT-5.4 (xhigh)	93.3% ±1.8

	Model	Score
1	GPT-5.2	49.8%
2	Claude Opus 4.5	45.5%
3	Claude Opus 4.1	43.6%
4	Claude Sonnet 4.5	42.5%
5	Gemini 3 Pro Preview	40.3%

	Model	Score
1	Claude Fable 5 (unknown)	1653.9
2	GLM 5.2 (max)	1593.3
3	Claude Opus 4.7 (unknown settings)	1566.8
4	Claude Opus 4.7 (no thinking)	1562.4
5	Claude Opus 4.6 (unknown settings)	1556.3

	Model	Score
1	Claude Opus 4.8 (xhigh)	47.1%
2	Claude Opus 4.7 (high)	44.1%
3	Claude Opus 4.6 (high)	41.2%
4	GPT-5.5 (xhigh)	40.2%
5	Claude Sonnet 5 (xhigh)	37.3%

	Model	Score
1	GPT-5 (medium)	97.2%
2	o3-pro	97.2%
3	Grok 4	94.4%
4	Grok 4 Fast	94.4%
5	Gemini 2.5 Pro Preview (Jun '25)	91.7%

	Model	Score
1	Gemini 3 Pro Preview	58.1% ±2.1
2	Gemini 3.1 Pro Preview	57.0% ±2.0
3	Gemini 3 Flash	48.1% ±2.4
4	Grok 4	43.6% ±2.2
5	Claude Opus 4.5	43.5% ±2.3

	Model	Score
1	GPT-5.5 Pro (xhigh)	100.0% ±0.0
2	GPT-5.5 (xhigh)	100.0% ±0.0
3	Claude Fable 5 (max)	99.7% ±0.3
4	Claude Opus 4.8	98.3% ±1.4
5	Claude Opus 4.7 (xhigh)	97.8% ±2.2

	Model	Score
1	GPT-5 (high)	98.1% ±0.3
2	GPT-5 (medium)	97.9% ±0.3
3	GPT-5 mini (high)	97.8% ±0.3
4	o4-mini (high)	97.8% ±0.3
5	o3 (high)	97.8% ±0.3

	Model	Score
1	GPT-5.5 Pro (xhigh)	87.7% ±1.9
2	Claude Fable 5 (max)	87.0% ±2.0
3	GPT-5.5 (xhigh)	85.3% ±2.1
4	GPT-5.4 Pro (xhigh)	82.5% ±2.3
5	Claude Opus 4.8 (max)	80.0% ±2.4

	Model	Score
1	Claude Fable 5 (max)	87.8% ±5.2
2	GPT-5.5 Pro (xhigh)	78.0% ±6.5
3	AI co-mathematician	75.6% ±6.7
4	GPT-5.5 (xhigh)	72.5% ±7.1
5	GPT-5.4 Pro (xhigh)	58.5% ±7.8

	Model	Score
1	Claude Fable 5 (max)	91.9%
2	Claude Opus 5 (max)	91.8%
3	Claude Opus 5 (high)	91.6%
4	GPT-5.6 Sol Pro (max)	89.4%
5	GPT-5.6 Sol (high)	88.8%

	Model	Score
1	Claude Opus 4.7 (unknown settings)	90.2%
2	GPT-5.5 (unknown thinking)	84.7%
3	GPT-5.4 (none)	81.8%
4	Gemini 3.1 Pro Preview	80.2%
5	Claude Opus 4.6 (unknown settings)	79.8%

	Model	Score
1	Gemini 3 Pro Preview	91.0%
2	GPT-5.2 (xhigh)	84.0%
3	Gemini 3 Flash	72.6%
4	GPT-5.2 (high)	67.0%
5	GPT-5 (high)	66.0%

	Model	Score
1	Gemini 3 Flash	88.0%
2	Gemini 2.5 Pro Preview (Jun '25)	86.0%
3	Gemini 3 Pro Preview	84.0%
4	Gemini 2.5 Pro Exp (Mar '25)	81.0%
5	GPT-5 (medium)	81.0%

AI Model Benchmarks

Compare Models

Humanity's Last Exam

SimpleBench

METR Time Horizons

SWE-bench Verified

GPQA Diamond

GDPval

Text Arena (Coding)

GSO (General Speedup Optimization)

Fiction.liveBench

BALROG

OTIS Mock AIME 2024-25

MATH Level 5

FrontierMath Tiers 1-3 (v2)

FrontierMath Tier 4 (v2)

WeirdML v2

Terminal-Bench 2.0

VPCT (Visual Physics Comprehension Test)

GeoBench