The Great Context Land Grab: 1M Tokens Becomes the New Standard

Week of October 14, 2026

The past seven days delivered a seismic shift in the API landscape. While the headlines will focus on Cohere's eyebrow-raising price hikes, the real story is buried in the metadata: the 1M context window has quietly become table stakes. From OpenAI's GPT-6 family to Anthropic's Claude 5.5 lineup, nearly every flagship model now ships with seven-figure context. Meanwhile, DeepSeek continues its race to the bottom on price—a strategy that should terrify anyone still paying 2024 rates.

The Main Event: Cohere's Pricing Pivot (and the Models That Just Got Cheaper)

Cohere Goes Premium—Aggressively So

If you're running Command R+ in production, we hope you're sitting down. Cohere raised prices by over 1,500% on its legacy Command R+ (08-2024) model, bringing input costs from $0.15 to $2.50 per 1M tokens and output from $0.60 to $10.00. The newer Command A+ saw a comparatively modest 733% input increase, landing at the same $2.50/$10.00 tier—though it did gain a 33% context bump to 256K tokens.

The takeaway: Cohere is clearly repositioning toward enterprise contracts and away from indie developers. If you're cost-sensitive, this is your signal to migrate.

DeepSeek Keeps Cutting—But the Real Story Is Elsewhere

DeepSeek V4 Pro 0423 shaved another 1.2% off both input ($0.4106) and output ($0.8213) costs. It's incremental, but it's relentless. The more interesting data point? DeepSeek's new Flash Latest variant clocks in at $0.02016 input / $0.891 output with a full 1M context window. That's not a typo.

This Week's Full Price & Context Changes

Price Movements

Cohere Command A+: Input ↑ 733.3% to $2.50/1M; Output ↑ 566.7% to $10.00/1M; Context ↑ 33.3% to 256K
Cohere Command R+ (08-2024): Input ↑ 1,566.7% to $2.50/1M; Output ↑ 1,566.7% to $10.00/1M
DeepSeek V4 Pro 0423: Input ↓ 1.2% to $0.4106/1M; Output ↓ 1.2% to $0.8213/1M

New Model Tracking (Selected Highlights)

The 1M Context Club:

OpenAI GPT-6 Sol / Sol Pro: $2/$10 per 1M (batch: $1/$5), 1.05M context
OpenAI GPT-6 Luna / Luna Pro: $0.10/$0.50 per 1M (batch: $0.05/$0.25), 1.05M context
OpenAI GPT-6 Astra / Astra Pro: $10/$50 per 1M (batch: $5/$25), 1.05M context
Anthropic Claude Sonnet 5.5: $2/$10 per 1M (batch: $1/$5), 1M context
Anthropic Claude Opus 5.5: $4/$20 per 1M (batch: $2/$10), 1M context
Anthropic Claude Haiku 5.5: $0.10/$0.50 per 1M (batch: $0.05/$0.25), 1M context
Google Gemini 4 Argon: $2/$10 per 1M (batch: $1/$5), 1.05M context
Mistral Large 4: $0.68/$2.09 per 1M, 1.05M context

Budget Disruptors:

ScaleDown Summarize / Extract: $0.04 input / free output, 1M context
DeepSeek Flash Latest: $0.02016/$0.891 per 1M, 1M context
OpenAI GPT-6 Luna: $0.10/$0.50 per 1M, 1.05M context
Anthropic Claude Haiku 5.5: $0.10/$0.50 per 1M, 1M context
Upstage Solar Mini 4: $0.05/$0.20 per 1M, 524K context
Nex AGI Nex-N2.5-Mini: $0.025/$0.10 per 1M, 262K context

Notable Entrants:

SpaceXAI Grok 4.7: $2/$6 per 1M, 500K context
MoonshotAI Kimi K3: $0.70/$15 per 1M, 1.05M context
Qwen3.8 Max Prime: $4/$12 per 1M, 1M context
Meta Muse Wonder: $4/$20 per 1M, 1.05M context
Sakana Fugu Ultra v2: $5/$30 per 1M, 1M context

Free Tier Expansions:

Dots3-Note Preview, LFM2.5-2.6B, Nemotron 3.5 Lightning, Gemma 4 31B, and multiple Poolside Laguna variants now offer $0 input/output.

What This Means for Your Stack

If you're on Cohere legacy models: Migrate. Yesterday.
If you need long context: The 1M window is now available at nearly every price point, from $0.02 (DeepSeek Flash) to $10 (GPT-6 Astra).
If you're building agents: The combination of 1M context + sub-$0.10 input pricing (GPT-6 Luna, Claude Haiku 5.5) makes multi-step reasoning economically viable at scale.
If you're watching batch pricing: Batch discounts remain a consistent 50%—worth architecting for if your workloads aren't latency-sensitive.

The compression continues. The question isn't whether prices will drop further—it's whether your architecture can keep up.

View live pricing and history at aimetricshub.cloud