Muse Spark Open 816B A42b added, Beam 1.0 added (17 Oct)

BenchLeader refreshed 1580 models from 24 of 24 data feeds (benchmark publishers, aggregators and price lists). 7 new models, 2 top-10 rank changes, 37 price changes, 78 new or updated benchmark results across 24 data feeds.

New models

Muse Spark Open 816B A42b (xhigh) (Meta) appeared with 5 benchmark results, entering the BenchLeader Index at #166. (details)
Beam 1.0 (Reflection) appeared with 7 benchmark results, entering the BenchLeader Index at #182. (details)
Nemotron 3.5 Super Vl 120B A12b Nvfp4 (NVIDIA) appeared with 3 benchmark results. (details)
Beam 1.0 501B A23B (Reflection) appeared with 2 benchmark results. (details)
Seed2 1 Pro (ByteDance) appeared with 1 benchmark result. (details)
DeepSeek V4.1 Flash 1017 (DeepSeek) is now listed at $0.26/M blended; no independent results yet. (details)
Salesforce Launa Omni 1 (Unknown) is now listed; no independent results yet. (details)

Movement in the top 10

GPT-6 Astra (high) moved from #5 to #4 in the BenchLeader Index. (details)
GPT-6 Astra (max) moved from #4 to #5 in the BenchLeader Index. (details)

Price changes

Nemotron 3 Ultra 550B A55B (thinking) output price moved from $2.50/M to $2.70/M (8% up). (details)
DeepSeek V4 Pro (high) input price moved from $0.44/M to $0.66/M (50% up). (details)
DeepSeek V4 Pro (high) output price moved from $0.87/M to $1.98/M (128% up). (details)
Muse Glimmer (high) cost per task moved from 0.1 to 0.1 (7% up). (details)
Nova 2.0 Pro Preview (medium) input price moved from $1.38/M to $1.25/M (9% down). (details)
Nova 2.0 Pro Preview (medium) output price moved from $11.00/M to $10.00/M (9% down). (details)
Gemma 4 26B A4B (thinking) output price moved from $0.40/M to $0.37/M (8% down). (details)
Qwen3-Next 80B A3B (thinking) input price moved from $0.15/M to $0.13/M (13% down). (details)
Nova 2.0 Pro Preview (low) input price moved from $1.38/M to $1.25/M (9% down). (details)
Nova 2.0 Pro Preview (low) output price moved from $11.00/M to $10.00/M (9% down). (details)
Nova 2.0 Lite (thinking) input price moved from $0.33/M to $0.30/M (9% down). (details)
Nova 2.0 Lite (thinking) output price moved from $2.75/M to $2.50/M (9% down). (details)
Nova 2.0 Lite (medium) input price moved from $0.33/M to $0.30/M (9% down). (details)
Nova 2.0 Lite (medium) output price moved from $2.75/M to $2.50/M (9% down). (details)
gpt-oss-120b (high) input price moved from $0.15/M to $0.13/M (13% down). (details)
gpt-oss-120b (high) cost per task moved from 0.1 to 0.1 (14% down). (details)
Qwen3-Next 80B A3B input price moved from $0.15/M to $0.13/M (13% down). (details)
DeepSeek V4 Pro (no reasoning) input price moved from $0.44/M to $1.32/M (200% up). (details)
DeepSeek V4 Pro (no reasoning) output price moved from $0.87/M to $3.96/M (355% up). (details)
Nova 2.0 Lite (low) input price moved from $0.33/M to $0.30/M (9% down). (details)
Nova 2.0 Lite (low) output price moved from $2.75/M to $2.50/M (9% down). (details)
gpt-oss-120b (low) input price moved from $0.15/M to $0.10/M (33% down). (details)
gpt-oss-120b (low) output price moved from $0.50/M to $0.59/M (18% up). (details)
gpt-oss-20b (high) input price moved from $0.07/M to $0.06/M (14% down). (details)
gpt-oss-20b (high) output price moved from $0.18/M to $0.22/M (22% up). (details)
gpt-oss-20b (high) cost per task moved from 0 to 0 (8% down). (details)
gpt-oss-20b (low) output price moved from $0.22/M to $0.25/M (14% up). (details)
Nova 2.0 Pro Preview (no reasoning) input price moved from $1.38/M to $1.25/M (9% down). (details)
Nova 2.0 Pro Preview (no reasoning) output price moved from $11.00/M to $10.00/M (9% down). (details)
Llama 4 Maverick output price moved from $0.87/M to $0.92/M (6% up). (details)
…and 7 more.

New benchmark results

Arena WebDev: 3 new results, including Kimi K2.5, DeepSeek V4 Pro, Devstral Medium (board)
AudioMC (text output): 2 new results, including Gemini 4 Argon, Gemini 3.8 Flash (board)
Chess Puzzles: 2 new results, including Mistral Large 4, GPT-6 Astra Pro (board)
Analyst Agent (AA): 1 new result, including Gemini 4 Argon (board)
FrontierCode 1.1 (AA): 1 new result, including Gemini 4 Argon (board)
Terminal-Bench-Science 0.1 (AA): 1 new result, including Gemini 4 Argon (board)
AA-3D: 1 new result, including Gemini 4 Argon (board)
Mystery Game Puzzles: 1 new result, including Muse Wonder (board)
PRBench Finance: 1 new result, including Gemini 4 Argon (board)
PRBench Legal: 1 new result, including Gemini 4 Argon (board)
HiL-Bench: 1 new result, including Gemini 4 Argon (board)
Vals Index: 1 new result, including Muse Spark Open (board)

Score revisions

Claude Opus 5.5 (max) on Arena Agent: 17.2 → 18.9. (details)
Gemini 4 Argon (high) on Arena Agent: 13.7 → 13.2. (details)
GPT-6.1 Sol (max) on Arena Agent: 14.8 → 13.6. (details)
GPT-6.1 Sol (max) on Arena Multi-turn: 1479 → 1486. (details)
Claude Opus 5 (max) on Arena Agent: 9.2 → 8.7. (details)
Claude Sonnet 5.5 (max) on Arena Agent: 12.5 → 13. (details)
Muse Wonder (max) on Arena Hard Prompts: 1534 → 1527. (details)
Muse Wonder (max) on Arena Coding: 1558 → 1548. (details)
Muse Wonder (max) on Arena Agent: 6.4 → 4.9. (details)
Muse Wonder (max) on Arena Instruction Following: 1503 → 1492. (details)
Muse Wonder (max) on Arena Multi-turn: 1544 → 1530. (details)
Claude Sonnet 5.5 (xhigh) on Arena WebDev: 1774 → 1768. (details)
Claude Sonnet 5.5 (xhigh) on Arena Maths: 1510 → 1502. (details)
GPT-6 Sol (max) on Arena Agent: 11.2 → 10.3. (details)
Qwen3.8 Max (max) on SciCode (AA): 53.2% → 52.1%. (details)
Qwen3.8 Max (max) on CritPt: 20% → 17.7%. (details)
Qwen3.8 Max (max) on τ³-Banking (AA): 51.3% → 47.8%. (details)
GLM 5.3 (max) on AA-3D: 782 → 787. (details)
Step 5 Preview on Arena Maths: 1471 → 1459. (details)
Gemini 3.1 Pro on Arena Agent: -3.5 → -4. (details)
…and 32 more.

Speed changes

GPT-5.5 Pro time to first token changed from 15.87 s to 3.19 s. (details)
Claude Fable 5.1 time to first answer changed from 3.05 s to 1.93 s. (details)
GLM 5.3 time to first token changed from 2.18 s to 0.89 s. (details)
GPT-5.4 Pro output speed changed from 5 tok/s to 9 tok/s. (details)
Gemini 3.7 Flash output speed changed from 285 tok/s to 80 tok/s. (details)
Claude Opus 4.6 time to first token changed from 1.03 s to 1.43 s. (details)
Gemini 3.5 Flash output speed changed from 188 tok/s to 114 tok/s. (details)
Step 5 Preview time to first token changed from 4.87 s to 2.13 s. (details)
Muse Spark Open output speed changed from 50 tok/s to 131 tok/s. (details)
Muse Spark Open time to first token changed from 1.74 s to 1.08 s. (details)
DeepSeek V4 Pro time to first answer changed from 48.01 s to 27.10 s. (details)
DeepSeek V4 Pro response time changed from 53.29 s to 33.44 s. (details)
DeepSeek V4 Pro thinking time changed from 46.19 s to 25.39 s. (details)
Claude Sonnet 5 time to first token changed from 1.25 s to 1.89 s. (details)
GPT-5.2-Codex time to first token changed from 1.72 s to 2.39 s. (details)
…and 65 more.

Today's top five

More: New AI model releases this week, and the full LLM leaderboard 2026.

This digest is generated automatically from the day's data changes. Every line links to the page where you can check the numbers and their source.