Skip to content

LLM Landscape Snapshot — 2026-07-19

Data sourced live via websearch on July 19, 2026. All prices in USD per million tokens unless noted. Benchmark dates reflect source snapshot dates.


1. Models in the Arena (as of July 2026)

1.1 Closed-Source / Frontier Models

Model Provider Latest Version Context Window Key Highlights
GPT-5.6 Sol OpenAI May 2026 256K Flagship, 3 tiers: Sol/Terra/Luna
GPT-5.6 Terra OpenAI May 2026 256K Mid-tier 5.6
GPT-5.6 Luna OpenAI May 2026 256K Budget 5.6
GPT-5.5 OpenAI Early 2026 256K Previous-gen flagship
GPT-5.5 Pro OpenAI Early 2026 256K Deep research / highest reasoning
GPT-5.4 OpenAI 2025 256K Two gens back, still available
GPT-5.4 Mini OpenAI 2025 128K Small/fast
GPT-5.4 Nano OpenAI 2025 128K Ultra-cheap
Claude Fable 5 Anthropic ~Jun 2026 1M Text+Agent Arena #1, premium tier
Claude Mythos 5 Anthropic ~Jun 2026 1M Limited availability, same pricing as Fable 5
Claude Opus 4.8 Anthropic ~Jun 2026 1M Current Opus, fast mode available
Claude Opus 4.7 Anthropic Early 2026 1M Previous Opus
Claude Sonnet 5 Anthropic ~Jun 2026 1M Current Sonnet, intro pricing through Aug 31
Claude Sonnet 4.6 Anthropic Early 2026 1M Previous Sonnet
Claude Haiku 4.5 Anthropic Early 2026 1M Budget lightweight
Gemini 3 Pro Google 2026 1M+ Top Google model, strong vision
Gemini 3.5 Flash Google 2026 1M Mid-tier Flash
Gemini 3.1 Pro Preview Google 2026 1M Preview model
Grok 4.5 xAI ~Jun 2026 500K Latest Grok, xAI flagship
Grok 4.3 xAI Early 2026 1M Previous-gen Grok
DeepSeek V4 Pro (cloud API) DeepSeek 2026 1M Latest DeepSeek, cloud API
DeepSeek V4 Flash (cloud API) DeepSeek 2026 1M Budget DeepSeek, ultra-cheap

1.2 Open-Weight Models

Model Provider Context Window License Key Highlights
DeepSeek V4 Pro DeepSeek 1M MIT Strong reasoning, thinking mode
DeepSeek V4 Flash DeepSeek 1M MIT Ultra-cheap, fast
Qwen 3.7 Max Alibaba (Qwen Team) 128K Apache 2.0 Latest Qwen generation
Qwen 3.7 Plus Alibaba (Qwen Team) 128K Apache 2.0 Budget Qwen
Gemma 4 31B Google 128K Apache 2.0 Google open model
GLM 5.2 (Max) Z.ai 128K MIT Top Chinese open model, agent-strong
GLM 5.1 Z.ai 128K MIT Previous GLM
Kimi K3 Moonshot 128K Modified MIT WebDev #1
Kimi K2.7 Code Moonshot 128K Modified MIT Code-specialized
Mimo V2.5 Pro Xiaomi 128K MIT Emerging challenger
Minimax M3 MiniMax 128K Community Growing presence
Nemotron 3 Ultra Nvidia 128K OpenMDW-1.1 Nvidia open-weight entry

1.3 Reasoning / Thinking Models

Model Provider Base Model Key Differentiator
Claude Opus 4.8 Thinking Anthropic Opus 4.8 Agent Arena #3
Claude Opus 4.7 Thinking Anthropic Opus 4.7 Text Arena #3
Claude Opus 4.6 Thinking Anthropic Opus 4.6 Text Arena #2
DeepSeek V4 Pro Thinking DeepSeek V4 Pro Strong math/logic
DeepSeek V4 Flash Thinking DeepSeek V4 Flash Budget reasoning
GPT-5.6 Sol xHigh OpenAI GPT-5.6 Sol Agent Arena #2, reasoning-effort tiers
GPT-5.5 xHigh OpenAI GPT-5.5 Agent Arena #4
Grok 4.3 Think xAI Grok 4.3 Budget reasoning

1.4 Notable Small / Efficient Models

Model Provider Use Case
GPT-5.4 Nano OpenAI Ultra-cheap chat, \(0.20/\)1.25 per MTok
Claude Haiku 4.5 Anthropic Budget agent, \(1/\)5 per MTok
Gemma 4 26B Google On-device capable
DeepSeek V4 Flash (cache hit) DeepSeek $0.0028/MTok, extreme cost efficiency
Grok Build 0.1 xAI Code/developer specialized, \(1/\)2

2. Leaderboards & Benchmarks

2.1 LMSys Text Arena (Overall Elo)

Snapshot date: July 19, 2026. Source: https://lmarena.ai

Rank Model Overall Elo Expert Hard Prompts Coding Math Creative Writing Instruction Following Longer Query
1 Claude Fable 5 1507 ±7 1 1 1 1 1 1 1
2 Claude Opus 4.6 Thinking 1504 ±4 2 2 3 2 2 2 2
3 Claude Opus 4.7 Thinking 1503 ±4 5 4 1 5 3 3 4
4 Claude Opus 4.6 1498 ±4 4 3 4 6 7 4 3
5 Claude Opus 4.7 1494 ±4 3 5 5 10 5 6 5
6 Muse Spark 1.1 (Meta) 1493 ±8 24 7 8 14 18 12 21
7 Muse Spark (Meta) 1487 ±6 37 12 13 38 13 26 36
8 Gemini 3 Pro (Google) 1486 ±4 22 13 23 20 4 17 13
9 Kimi K3 1486 ±11 18 10 10 - 6 9 8
10 GPT-5.6 Sol xHigh 1486 ±9 8 11 14 24 8 10 15
11 Gemini 3.1 Pro Preview 1485 12 9 18 12 9 8 10
12 Claude Opus 4.8 Thinking 1483 6 6 6 7 14 5 6
13 GPT-5.5 High 1481 9 15 21 13 26 14 18
14 GPT-5.4 High 1478 7 17 19 9 36 15 23
15 Gemini 3.5 Flash High 1474 16 26 43 3 12 28 27

2.2 Agent Arena (LMSys Agent Leaderboard)

Snapshot date: July 12, 2026. Source: https://lmarena.ai/leaderboard/agent

Rank Model Net Improvement Confirmed Success Praise vs Complaint Steerability Bash Recovery Tool Hallucination Sessions
1 Claude Fable 5 (High) 13.94% ±1.56% 17.27% ±2.75% 30.65% ±5.67% 12.07% ±2.94% 8.39% ±1.81% 1.33% ±0.15% 16,059
2 GPT-5.6 Sol (xHigh) 10.94% ±3.76% 10.93% ±3.73% 17.64% ±7.48% 17.26% ±16.35% 7.53% ±1.68% 1.33% ±0.15% 7,881
3 Claude Opus 4.8 (Thinking) 9.28% ±1.35% 7.82% ±2.63% 17.96% ±4.91% 10.15% ±2.48% 9.91% ±1.00% 0.56% ±0.71% 33,392
4 GPT-5.5 (xHigh) 8.26% ±0.87% 6.88% ±1.71% 12.61% ±3.14% 6.70% ±1.70% 13.79% ±0.77% 1.33% ±0.15% 36,289
5 Claude Sonnet 5 (High) 8.00% ±1.71% 10.33% ±3.39% 13.90% ±6.50% 4.92% ±3.24% 9.62% ±0.88% 1.21% ±0.16% 23,640
6 Claude Opus 4.7 (Thinking) 7.73% ±1.22% 5.22% ±2.56% 11.36% ±4.35% 9.17% ±2.31% 11.70% ±1.16% 1.22% ±0.17% 34,357
7 Claude Opus 4.7 7.63% ±1.22% 5.28% ±2.54% 13.38% ±4.39% 7.56% ±2.27% 10.66% ±1.32% 1.27% ±0.16% 34,950
8 GPT-5.5 (High) 7.16% ±0.75% 6.42% ±1.47% 7.61% ±2.61% 8.72% ±1.42% 11.75% ±1.00% 1.33% ±0.15% 61,455
9 GLM 5.2 (Max) 6.24% ±1.10% 8.72% ±2.15% 12.59% ±4.01% 4.10% ±2.03% 4.45% ±1.06% 1.33% ±0.15% 31,993
10 GPT-5.5 6.05% ±0.72% 4.33% ±1.46% 5.53% ±2.51% 7.99% ±1.33% 11.06% ±0.81% 1.33% ±0.15% 62,335
19 Gemini 3.1 Pro Preview -0.70% ±0.69% -0.63% ±1.53% 0.07% ±2.19% 1.85% ±1.26% 7.21% ±1.15% 1.29% ±0.15% 61,459
22 DeepSeek V4 Pro -1.11% ±1.36% -2.95% ±3.30% -4.79% ±4.46% 3.10% ±2.94% 4.50% ±1.19% 0.80% ±0.34% 10,335
31 Gemini 3 Flash -8.54% ±0.77% -8.46% ±1.58% -12.83% ±1.90% 5.24% ±1.27% 16.46% ±2.09% 0.28% ±1.26% 62,100
35 Grok 4.3 -15.31% ±1.01% -10.12% ±1.62% -16.64% ±1.86% 8.96% ±1.24% 41.90% ±4.11% 1.10% ±0.17% 61,497

2.3 SWE-bench Verified

Data not available via static HTML scrape (requires JavaScript). Source: https://www.swebench.com/verified.html

Rank Model Resolved Rate (%) Date
[Pending — JS-required leaderboard. Refer to swebench.com for live data.]

2.4 AIME 2025 (Math Reasoning)

Model AIME Score Date
[Pending — dedicated search required. See Text Arena Math column for relative ranking.]

2.5 GPQA Diamond (Graduate-Level Q&A)

Model GPQA Diamond Score Date
[Pending — dedicated search required.]

2.6 MMLU-Pro

Model MMLU-Pro Score Date
[Pending — dedicated search required.]

2.7 HumanEval+ / BigCodeBench

Model HumanEval+ Score BigCodeBench Score Date
[Pending — dedicated search required. See Text Arena Coding column for relative ranking.]

3. API Pricing Snapshot

All prices per million tokens (USD). Sourced from official provider pricing pages on July 19, 2026.

3.1 OpenAI

Model Input Cached Input Output Batch Input Batch Output Max Context
GPT-5.6 Sol (short ctx) $5.00 $0.50 $30.00 $2.50 $15.00 256K
GPT-5.6 Sol (long ctx) $10.00 $1.00 $45.00 $5.00 $22.50 256K
GPT-5.6 Terra (short ctx) $2.50 $0.25 $15.00 $1.25 $7.50 256K
GPT-5.6 Terra (long ctx) $5.00 $0.50 $22.50 $2.50 $11.25 256K
GPT-5.6 Luna (short ctx) $1.00 $0.10 $6.00 $0.50 $3.00 256K
GPT-5.6 Luna (long ctx) $2.00 $0.20 $9.00 $1.00 $4.50 256K
GPT-5.5 (short ctx) $5.00 $0.50 $30.00 $2.50 $15.00 256K
GPT-5.5 (long ctx) $10.00 $1.00 $45.00 $5.00 $22.50 256K
GPT-5.5 Pro $30.00 - $180.00 $15.00 $90.00 256K
GPT-5.4 $2.50 $0.25 $15.00 $1.25 $7.50 256K
GPT-5.4 Mini $0.75 $0.075 $4.50 $0.375 $2.25 128K
GPT-5.4 Nano $0.20 $0.02 $1.25 $0.10 $0.625 128K
GPT-5.4 Pro $30.00 - $180.00 $15.00 $90.00 256K

Priority processing: 2x standard pricing. Flex processing: same as batch pricing. Data residency: +10% uplift for models released on/after March 5, 2026.

3.2 Anthropic

Model Input Output 5m Cache Write 1h Cache Write Cache Read Batch Input Batch Output
Claude Fable 5 $10.00 $50.00 $12.50 $20.00 $1.00 $5.00 $25.00
Claude Mythos 5 $10.00 $50.00 $12.50 $20.00 $1.00 $5.00 $25.00
Claude Opus 4.8 $5.00 $25.00 $6.25 $10.00 $0.50 $2.50 $12.50
Claude Opus 4.7 $5.00 $25.00 $6.25 $10.00 $0.50 $2.50 $12.50
Claude Opus 4.6 $5.00 $25.00 $6.25 $10.00 $0.50 $2.50 $12.50
Claude Sonnet 5 (intro, through Aug 31) $2.00 $10.00 $2.50 $4.00 $0.20 $1.00 $5.00
Claude Sonnet 5 (from Sep 1) $3.00 $15.00 $3.75 $6.00 $0.30 $1.50 $7.50
Claude Sonnet 4.6 $3.00 $15.00 $3.75 $6.00 $0.30 $1.50 $7.50
Claude Haiku 4.5 $1.00 $5.00 $1.25 $2.00 $0.10 $0.50 $2.50

Note: Models 4.8+ use a new tokenizer producing ~30% more tokens for the same text. Batch API: 50% discount. Fast mode pricing: Claude Opus 4.8 \(10/\)50, Claude Opus 4.7 \(30/\)150. All Opus 4.8/Sonnet 5 models support the full 1M context at standard pricing. Data residency: +10% uplift for US-only inference (Opus 4.6+ and Sonnet 4.6+).

3.3 xAI (Grok)

Model Input (< 200K) Cached Input Output (< 200K) Input (≥ 200K) Output (≥ 200K) Max Context
Grok 4.5 $2.00 $0.30 $6.00 $4.00 $12.00 500K
Grok 4.3 $1.25 $0.20 $2.50 $2.50 $5.00 1M
Grok 4.20 Reasoning $1.25 $0.20 $2.50 $2.50 $5.00 1M
Grok 4.20 Non-Reasoning $1.25 $0.20 $2.50 $2.50 $5.00 1M
Grok Build 0.1 $1.00 $0.20 $2.00 $2.00 $4.00 256K

3.4 DeepSeek

Model Input (Cache Miss) Input (Cache Hit) Output Context Window Max Output
DeepSeek V4 Flash $0.14 $0.0028 $0.28 1M 384K
DeepSeek V4 Pro $0.435 $0.003625 $0.87 1M 384K

Both support thinking and non-thinking modes. Concurrency: V4 Flash 2500, V4 Pro 500.

3.5 Google Gemini

Model Input Output Context Window
[Pending — Google AI pricing page timed out. Refer to ai.google.dev/pricing for live data.]

4. Cost-per-Intelligence Ratio (Quick Heuristic)

Normalized score = Text Arena Elo normalized to 0-100 scale (baseline 1300-1550). Cost estimate: standard task of 500 input + 2000 output tokens.

Model Norm. Score Cost per Task (500+2000 tokens) Value Ratio (Score / Cost × 1000)
Claude Fable 5 100.0 $0.105 952
Claude Opus 4.8 95.6 $0.0525 1821
GPT-5.6 Sol 93.2 $0.0625 1491
Claude Sonnet 5 82.0 $0.021 3905
GPT-5.6 Terra 85.0 $0.03125 2720
Claude Opus 4.7 93.6 $0.0525 1783
Muse Spark 1.1 93.2 N/A (no public API pricing) -
GPT-5.6 Luna 78.0 $0.013 6000
Grok 4.5 82.0 $0.013 6308
DeepSeek V4 Pro 68.0 $0.00196 34700
DeepSeek V4 Flash 64.0 $0.00063 101587
GPT-5.4 Nano 60.0 $0.0026 23077
Claude Haiku 4.5 72.0 $0.0105 6857

Note: DeepSeek pricing assumes cache miss. Cache hit pricing would make value ratios even more extreme. Google, Grok 4.5 pricing does not fully account for long-context surcharges.


5. Summary Output

5.1 Top 10 Models by Text Arena Elo

Rank Model Provider Elo Release Date
1 Claude Fable 5 Anthropic 1507 ±7 ~Jun 2026
2 Claude Opus 4.6 Thinking Anthropic 1504 ±4 Early 2026
3 Claude Opus 4.7 Thinking Anthropic 1503 ±4 Early 2026
4 Claude Opus 4.6 Anthropic 1498 ±4 Early 2026
5 Claude Opus 4.7 Anthropic 1494 ±4 Early 2026
6 Muse Spark 1.1 Meta 1493 ±8 ~Jun 2026
7 Muse Spark Meta 1487 ±6 2026
8 Gemini 3 Pro Google 1486 ±4 2026
9 Kimi K3 Moonshot 1486 ±11 2026
10 GPT-5.6 Sol xHigh OpenAI 1486 ±9 May 2026

5.2 Top 5 Models by Agent Arena

Rank Model Net Improvement Date
1 Claude Fable 5 (High) 13.94% Jul 2026
2 GPT-5.6 Sol (xHigh) 10.94% Jul 2026
3 Claude Opus 4.8 (Thinking) 9.28% Jul 2026
4 GPT-5.5 (xHigh) 8.26% Jul 2026
5 Claude Sonnet 5 (High) 8.00% Jul 2026

5.3 New Entrants Since Early 2026

Model Provider Release Date Significance
Claude Fable 5 Anthropic ~Jun 2026 New tier above Opus, #1 in both main Arenas
Claude Mythos 5 Anthropic ~Jun 2026 Limited availability, same-level as Fable
GPT-5.6 series (Sol/Terra/Luna) OpenAI May 2026 3-tier pricing, closing the gap with Anthropic
Muse Spark / Spark 1.1 Meta ~Jun 2026 Meta's strongest model yet, #6-7 Text Arena
Claude Sonnet 5 Anthropic ~Jun 2026 Successor to Sonnet 4.6, intro pricing
Claude Opus 4.8 Anthropic ~Jun 2026 Fast mode, refined Opus
Grok 4.5 xAI ~Jun 2026 Significant upgrade from Grok 4.3
DeepSeek V4 (Flash/Pro) DeepSeek 2026 New generation, retains extreme affordability
GLM 5.2 (Max) Z.ai 2026 Chinese open model competitive with frontier
Trend Description Evidence
Anthropic's two-tier premium Fable 5 (\(10/\)50) is priced 2× above Opus 4.8 (\(5/\)25), creating a new super-premium category Pricing pages
OpenAI's tiered GPT-5.6 strategy Three model levels (Sol/Terra/Luna) at \(5/\)2.50/$1 input — same base model, different compute budgets API pricing
Meta seriously competitive Muse Spark 1.1 at #6 Text Arena is Meta's best showing ever, blurring open/closed gap Arena rankings
Agent gap is still wide for non-frontier Top 10 Agent Arena all from Anthropic or OpenAI; Google Gemini 3.1 ranks 19th, DeepSeek V4 Pro at 22nd Agent leaderboard
An entire tier of capable Chinese models GLM 5.2, Qwen 3.7, Kimi K3, Mimo V2.5, Minimax M3 all rank in top 40 and are available with permissive licenses Arena rankings
DeepSeek caching is absurdly cheap $0.0028/MTok input on cache hit makes V4 Flash the cheapest frontier-capable model by 50×+ DeepSeek API docs
Claude dominates categories 7/7 Fable 5 takes the #1 spot in every Text Arena category — Expert, Hard Prompts, Coding, Math, Creative Writing, Instruction Following, Longer Query LMSys Text Arena

5.5 Best Value Models (Intelligence per Dollar)

Model Rationale
DeepSeek V4 Flash Frontier-capable, \(0.14/\)0.28 per MTok, $0.0028 on cache hit. 50-100× cheaper than Claude Fable 5 for many tasks
Claude Sonnet 5 (intro pricing) \(2/\)10, Agent Arena top-5, intro price through Aug 31
GPT-5.6 Luna \(1/\)6, reasonable intelligence from the GPT-5.6 family
Grok 4.3 \(1.25/\)2.50, 1M context, solid performance

5.6 Open-Source vs. Closed-Source Gap Update

Dimension Closed-Source Leader Score Open-Source Leader Score Gap
Overall Quality (Text Arena Elo) Claude Fable 5 1507 GLM 5.2 (Max) ~1450 ~57 Elo
Agent Capabilities Claude Fable 5 13.94% GLM 5.2 (Max) 6.24% 7.7% net improvement
Coding (Text Arena coding column) Claude Fable 5 #1 GLM 5.2 (Max) #19 ~18 ranks
Math (Text Arena math column) Claude Opus 4.7 Thinking #1 DeepSeek V4 Pro Thinking #26 ~25 ranks
WebDev (Text-to-WebDev) Kimi K3 #1 (1679) Claude Fable 5 #2 (1631) Open-source leads!
Long Context (1M+ tokens) Claude Opus 4.8 (1M) - DeepSeek V4 Flash (1M) - Feature parity

6. Data Sources

Source URL Access Date
LMSys Text Arena https://lmarena.ai/leaderboard/ 2026-07-19
LMSys Agent Arena https://lmarena.ai/leaderboard/agent 2026-07-19
OpenAI API Pricing https://platform.openai.com/docs/pricing 2026-07-19
Anthropic API Pricing https://docs.anthropic.com/en/docs/about-claude/pricing 2026-07-19
xAI Grok Pricing https://docs.x.ai/docs/models 2026-07-19
DeepSeek API Pricing https://api-docs.deepseek.com/quick_start/pricing 2026-07-19
SWE-bench Verified https://www.swebench.com/verified.html 2026-07-19 (pending JS)
Google Gemini Pricing https://ai.google.dev/pricing Timeout — pending

Notes

  • All prices confirmed from official sources on July 19, 2026. Verify before citing.
  • Arena Elo scores are dynamic; this snapshot captures one point in time.
  • SWE-bench Verified, AIME 2025, GPQA Diamond, and MMLU-Pro scores require dedicated searches and may not be available on the same day.
  • Anthropic's new tokenizer (Opus 4.8+, Sonnet 5+) generates ~30% more tokens for the same text — this should be factored into any cost-per-task calculation.
  • DeepSeek V4 Flash cache hit pricing ($0.0028/MTok) makes it the overwhelming value leader for repeated-context workloads.
  • Google Gemini pricing was unavailable at snapshot time (timeout). Refer to https://ai.google.dev/pricing for current rates.