LLM Best Value Chart

As of Oct 29, 5 models sit on the price–performance frontier.

Top agentClaude Opus 5.520.94% Top in TextGemini 4 Carbon1,527.5 Intelligence #1Gemini 4 Carbon59.4 Best valueMiMo V2.5$0.27/1M out Open weights #1Kimi K3Arena 1,488.2 Open · top Intel.Kimi K3.1 Pro46.0 · Agent Arena #10, no Text Arena score yet

This week

vs Oct 23
  1. Qwen3.8 MaxAlibaba▼5#23 → #28
  2. Muse Spark 1.1Meta▼4#40 → #44
  3. GPT-6 AstraOpenAI▼3#2 → #5
  4. Claude Sonnet 5Anthropic▼3#11 → #14
  5. Claude Haiku 5.5Anthropic▲3#24 → #21

New models

last 14 days
  1. Gemini 4 CarbonGoogleOct 28
  2. GPT-6.1 TerraOpenAINo Arena yetOct 26
  3. Kimi K3.1 ProMoonshot AIOct 23
Showing 18 modelsPrice unit: USD / million tokens

Scores and prices at a glance

0 20 40 60 $1.00 $10.00 $100 Intelligence Index Output price · USD / million tokens (log scale) G Gemini 4 Carbon · Speed 88/s G Gemini 4 Argon · Speed 80/s A Claude Opus 5.5 · Speed 96/s · Effort tiers 34.5–58.4 M Muse Wonder · Speed 181/s A Claude Fable 5.1 · Speed 67/s · Effort tiers 38.5–49.7 G Gemini 3.8 Flash · Speed 133/s · Effort tiers 22.3–30.5 M Muse Spark 1.3 A Qwen3.8 Max · Speed 39/s O GPT-6.1 Sol · Speed 80/s · Effort tiers 34.5–54.5 Z GLM-5.3 · Speed 77/s · Effort tiers 23.1–31.7 A Claude Sonnet 5.5 · Speed 137/s · Effort tiers 25.3–54.2 O GPT-6 Astra · Speed 70/s · Effort tiers 43.0–56.9 D DeepSeek V4 Pro · Speed 76/s x Grok 4.7 · Speed 83/s · Effort tiers 30.9–35.0 M MiniMax M3 · Speed 104/s A Claude Haiku 5.5 · Speed 236/s · Effort tiers 18.6–36.2 M Kimi K3.1 Pro · Speed 33/s O GPT-6.1 Terra · Speed 91/s · Effort tiers 29.2–51.0
Price-score frontierBubble size = output speed 33–236/s

How to read the chart

Higher means a better score; farther left means cheaper output. Select an icon for details.

The green line uses all models with a score and an output price above zero; filters do not change this reference line.

Free, unpriced, or unscored models are excluded from the log-scale price chart; see the rankings instead.

Scoring method and data sources

Ranked by LMArena Arena score by default, switchable to the Artificial Analysis Intelligence Index; Arena is not an independent coding evaluation. Prices are official API list prices, with OpenRouter quotes shown separately as the lowest route price. Open weights do not mean unrestricted commercial use. Data sources

Models in the chart
  1. Claude Opus 5.558.4$20.00
  2. Gemini 4 Carbon59.4$20.00
  3. Claude Sonnet 5.554.2$10.00
  4. Claude Fable 5.149.0$50.00
  5. GPT-6 Astra56.9$50.00
  6. GPT-6.1 Sol54.5$10.00
  7. Gemini 4 Argon50.4$20.00
  8. Kimi K3.1 Pro46.0$10.00
  9. Muse Wonder43.3$20.00
  10. Gemini 3.8 Flash30.5$3.75
  11. Muse Spark 1.336.4$4.25
  12. Grok 4.735.0$6.00
  13. Claude Haiku 5.536.2$0.50
  14. GLM-5.331.7$4.40
  15. Qwen3.8 Max33.6$6.00
  16. DeepSeek V4 Pro23.1$3.96
  17. MiniMax M318.7$1.20
  18. GPT-6.1 Terra51.0$5.00

Not yet ranked by Arena 5

Tracked models that have no LMArena or Artificial Analysis score yet, newest first. They move into the table once scored.

Sources & methodology

As of Oct 29, 2026, 5 models sit on the price–performance frontier: nothing is stronger at the same price. The cheapest frontier model is Claude Haiku 5.5 ($0.50 per 1M output tokens); the strongest is Gemini 4 Carbon (score 59.4, $20.00 output).

  • arena.ai Agent Arena

    Agent Arena (Capabilities · Overall): Net Improvement, Confirmed Success, Praise vs Complaint and Steerability with 95% CIs, sessions and ranks (arena.ai, via the public Hugging Face dataset lmarena-ai/leaderboard-dataset, configs agent / agent_task_outcome_explicit / agent_praise_complaint / agent_steerability, CC BY 4.0). Hugging Face

  • LMArena

    Arena score, 95% confidence interval, votes and rank (overall Text leaderboard with style control, the same view arena.ai shows by default; dataset CC BY 4.0).

  • Artificial Analysis

    Intelligence Index, Coding Index, output speed, TTFT / first-answer latency, release dates and official prices (shown on this page only, not in the public API).

  • OpenRouter

    Per-provider prices (the cheapest becomes “lowest route”), context length and listing dates.

  • Vendor pricing pages

    “Input / Output” are official API list prices (USD per 1M tokens); when no official price exists we use the OpenRouter price and mark it with *.

How we compute it

  1. Ranking: by default identical to arena.ai Agent Arena (Capabilities · Overall) — sorted by Net Improvement, using arena's own rank. Rank spread comes from 95% CIs: best = 1 + models whose lower bound exceeds this model's upper bound; worst = 1 + other models whose upper bound exceeds this model's lower bound. Models missing from the public dataset (mostly ones arena marks Preliminary) are hand-entered from the arena.ai page with the entry date — only values visible on the page, nothing estimated; once the dataset includes them it takes over, and entries older than 21 days raise a warning. Models without agent results follow, sorted by Text Arena score (style control; "#n" is the text rank).
  2. Rank: by default, LMArena Text Arena score with style control (the same view arena.ai shows by default; Bradley-Terry, not Elo). The ± is the half-width of the 95% confidence interval, and neighbouring ranks often overlap. Models with fewer than 5,000 votes or ± above 10 are marked Preliminary. We do not invent a composite score.
  3. New models: models released within 30 days (up to 6) that have an AA Intelligence Index but no Arena Text or Agents score are shown in a “New · no Arena score yet” group inside the top 30 rows of the default view, ordered by AA Intelligence Index and marked “New” without taking a rank; they move into the ranking automatically once Arena lists them. Models without even an AA index are listed under “Not ranked yet” at the bottom. Vendor names are merged (e.g. Kimi → Moonshot AI); open-weight status comes from upstream metadata, gaps are filled only from a sourced override list, and unknowns show –.
  4. Intelligence / Coding Index: Artificial Analysis aggregates of public benchmarks. When a model has several reasoning-effort tiers, the table shows the best tier for the current sort column; switch to “by tier” to see each one.
  5. Prices are USD per 1M tokens. “Lowest route” is the cheapest OpenRouter provider’s input / output price; ⚠ marks routes more than 60% below list (usually promos or a single provider). The best-value board only includes models with Arena ≥ 1400 or Intelligence Index ≥ 40, sorted by official output price.
  6. Latency and speed: TTFT is the time to the first token (including thinking tokens); “Answer” is the time to the first answer token (includes the thinking phase for reasoning models); speed is output tokens per second. All are Artificial Analysis medians.
  7. Updates: Text and other data sync daily at 14:20 (UTC+8), the Agents board at 14:25. “This week” compares against the daily record from 7 days earlier (or the oldest one, with its age shown, while fewer than 7 days exist); rank arrows in the table appear after 14 days of records.
  8. Updates: synced automatically every day at 14:20 (UTC+8). LMArena publishes on its own schedule, so the “Arena published” date can be older than the update date. “This week” compares against the daily archive from 7 days earlier.
  9. Open data: /api/public/llm-leaderboard.php (JSON, LMArena and OpenRouter fields only).

FAQ

What is the best AI model right now?

As of Oct 29, 2026, the arena.ai Agents leaderboard (Net Improvement in real agent sessions) is led by Claude Opus 5.5 (20.94%, 95% interval 18.79%–23.09%), ahead of Gemini 4 Carbon (16.04%). The top intervals overlap heavily, so the exact order is not reliable. On the LMArena Text leaderboard, Gemini 4 Carbon leads (1,527.5).

Which good AI model is the cheapest?

Among models above the quality floor (Arena ≥ 1400 or Intelligence Index ≥ 40), MiMo V2.5 has the lowest official price: $0.13 input / $0.27 output per 1M tokens, followed by DeepSeek V4 Flash ($0.28 output). See the best-value leaderboard.

What is the best open-source (open-weight) AI model?

By Arena score, the top open-weight model is Kimi K3 (Moonshot AI, 1,488.2), followed by GLM-5.3 (1,480.6). We track 18 open-weight models with an Arena score — see the open-weight leaderboard. Open weights do not mean unrestricted commercial use; check each licence. By Artificial Analysis Intelligence Index, the top open-weights model is Kimi K3.1 Pro (Moonshot AI, 46.0, Agent Arena #10, no Text Arena score yet).

Which AI model is the fastest?

The fastest measured output speed belongs to Claude Haiku 5.5 (236 tokens/s); the lowest time to first answer token belongs to Grok 4.7 (10.25 s). These are Artificial Analysis medians; for reasoning models the first-answer latency includes thinking time.

Which AI models were released recently?

3 models released in the last 14 days are listed, including: Gemini 4 Carbon (Oct 28, 2026), GPT-6.1 Terra (Oct 26, 2026), Kimi K3.1 Pro (Oct 23, 2026). New models usually take days to weeks to enter the LMArena ranking; until then, any model with an Artificial Analysis Intelligence Index is shown in the “New · no Arena score yet” group within the top 30 rows of the default view, ordered by that index and marked “New” without taking a rank.

How often is the data updated, and where does it come from?

Synced automatically every day: the Agents board comes from arena.ai (public Hugging Face dataset, current release Oct 28, 2026), Text scores from LMArena (released Oct 28, 2026); intelligence and coding indices, speed and latency from Artificial Analysis; route prices and context from OpenRouter. This page was updated Oct 29, 2026.

Why do prices differ from the official site?

Input/output prices are official list prices in USD per million tokens. The lowest route price is the cheapest OpenRouter route, which may be a promotion or a single provider and can differ from local resellers.